The applications and prospects of big data in perioperative anesthetic management

Abstract

Perioperative anesthetic management entails a multitude of decision-making processes within complex medical scenarios. These demand the continuous and dynamic execution of precise decisions which poses significant challenges. In the age of big data, the exponential growth in data volume from diverse sources has revolutionized many fields, including healthcare, finance, and marketing. Machine learning has emerged as a powerful tool for analyzing big data, enabling the handling of large datasets and uncovering intricate patterns and relationships. The application of big data and artificial intelligence algorithms is gradually being integrated, enabling effective task completion in various stages of perioperative management, including risk prediction, decision support, and auxiliary examination. Through in-depth analysis of big data, healthcare professionals can gain insights into patient prognoses. This review provides a comprehensive overview of the distinctive features of perioperative big data and its applications in anesthesia management during the perioperative period.

Full text

Zhuetal.
Anesthesiology and Perioperative Science (2024) 2:30
https://doi.org/10.1007/s44254-024-00068-0
REVIEW ARTICLE Open Access
© The Author(s) 2024. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which
permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the
original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or
other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line
to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory
regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this
licence, visit http://creativecommons.org/licenses/by/4.0/.
Anesthesiology and
Perioperative Scienc
e
The applications andprospects ofbig data
inperioperative anesthetic management
Yiziting Zhu1†, Xiang Liu1†, Yujie Li1 and Bin Yi1*
Abstract
Perioperative anesthetic management entails a multitude of decision-making processes within complex medical sce-
narios. These demand the continuous and dynamic execution of precise decisions which poses significant challenges.
In the age of big data, the exponential growth in data volume from diverse sources has revolutionized many fields,
including healthcare, finance, and marketing. Machine learning has emerged as a powerful tool for analyzing big data,
enabling the handling of large datasets and uncovering intricate patterns and relationships. The application of big
data and artificial intelligence algorithms is gradually being integrated, enabling effective task completion in various
stages of perioperative management, including risk prediction, decision support, and auxiliary examination. Through
in-depth analysis of big data, healthcare professionals can gain insights into patient prognoses. This review provides
a comprehensive overview of the distinctive features of perioperative big data and its applications in anesthesia man-
agement during the perioperative period.
Keywords Big data, Machine learning, Perioperative management, Prediction
1 Introduction
As computer processing capabilities have advanced and
information technology has undergone significant devel-
opment, the volume of data generated from diverse
sources has experienced exponential growth [1]. e
characteristics of big data can be defined in the form of
‘4V’, meaning volume (large volume), variety (different
methods), velocity (fast production), and value (high
quality with valuable insights) [2]. Consequently, big data
poses inherent challenges to conventional data process-
ing methodologies, mandating the adoption of special-
ized tools and analytical techniques for comprehensive
exploration. Machine learning is now commonly used for
analyzing big data because it can handle large datasets at
scale and automatically learn patterns and relationships
in the data without being explicitly programmed [1].
Moreover, it performs better in complex and non-linear
relationships than conventional statistical methods [3–5].
Perioperative management is a pivotal aspect of health-
care encompassing the comprehensive care provided to
patients in the preoperative, intraoperative, and post-
operative phases of surgery. Reasonable perioperative
management can reduce complications and postopera-
tive length of stay along with risks of readmission and
mortality [6, 7]; these have a profound effect on patients’
convalescence and overall well-being. Perioperative man-
agement encompasses a comprehensive continuum that
spans the preoperative, intraoperative, and postopera-
tive phases of medical care, necessitating multidiscipli-
nary collaboration taking account of individual patient
variations and the systematic evaluation of associated
risks. It is characterized by personalized management,
interdisciplinary cooperation, temporal constraints,
and the integration of multifarious information sources.
Notably, perioperative anesthesia management requires
anesthesiologists to continually formulate adaptive deci-
sions, drawing upon real-time monitoring data, clinical
†Yiziting Zhu and Xiang Liu contributed equally to this work.
*Correspondence:
Bin Yi
yibin1974@163.com
1 Department of Anesthesiology, Southwest Hospital, Third Military
Medical University, Chongqing 400037, China

Page 2 of 11
Zhuetal. Anesthesiology and Perioperative Science (2024) 2:30
signs and surgical progress, to ensure patient safety and
surgical success. ese tasks in the dynamic periopera-
tive environment presents a formidable challenge. Con-
sequently, effective perioperative management relies on
the support of perioperative big data to realize personal-
ized decision-making, interdisciplinary coordination, and
dynamic adaptation.
rough the comprehensive analysis of big data, doc-
tors can derive insights into patient prognoses, identify
potential areas for enhancement, and develop more effec-
tive treatment strategies, thereby enhancing the overall
quality of patient care. is review summarizes the appli-
cation of big data in the perioperative period and deline-
ates prospects in this evolving field.
2 Characteristics ofperioperative big data
e perioperative period encompasses the entirety of
the surgical journey, commencing with the decision to
pursue a surgical intervention and concluding with the
patient’s complete recuperation. Perioperative big data
refers to the massive, multivariate, high-frequency and
multi-dimensional medical data collected in this process.
e focus of perioperative management varies across
the distinct phases of the perioperative period. In concert
with the advancement of medical information technol-
ogy, each phase of this perioperative continuum engen-
ders a substantial volume of data. ese perioperative
data originate from diverse sources, spanning multiple
data platforms such as hospital information systems and
other datasets, as shown in Fig. 1. Within these data-
sets, structured and unstructured data coexist, contain-
ing multi-dimensional information including patient
demographics, medical history, laboratory test results,
imageological examinations, preoperative assessments,
intraoperative monitoring data, anesthesia-related infor-
mation, surgical details, and postoperative recovery met-
rics. Moreover, these data arise from various medical
devices and systems, often characterized by incongruent
data formats and standards.
Furthermore, a significant interrelationship exists
among perioperative data elements, such as the
Fig.1 Sources of perioperative big data

Page 3 of 11
Zhuetal. Anesthesiology and Perioperative Science (2024) 2:30
underlying disease of the patient, the results of the pre-
operative examination and the complexity and risk asso-
ciated with the surgical procedure. Consequently, the
attributes of perioperative big data are typified by high
volume, diversity, high dimensionality, correlation, and
heterogeneity. Due to the diversity of surgery and anes-
thesia protocols, instances of missing values or vacant
data fields, referred to as ’data sparsity’, are commonplace
within these datasets. During intraoperative anesthesia
management, anesthesiologists are tasked with real-time
monitoring of multiple physiological parameters while
making consequential decisions. Continuous data gen-
eration stems from monitors, anesthesia apparatus, and
infusion pumps. Additionally, the data collection process
may introduce noise due to human errors, signal interfer-
ence, and imprecise equipment calibration. erefore,
perioperative big data is also characterized as dynamic,
temporal and with high levels of noise. Given the sensi-
tive nature of medical data, which frequently contains
private and confidential patient information, meticulous
attention must be devoted to ensuring secure data pro-
cessing, storage, and sharing.
Generally, perioperative big data exhibits character-
istics of high dimensionality, multimodality, dynamism,
strong correlations, significant mixed noise, and frequent
specific missing data [8]. ese characteristics not only
present challenges in terms of data storage, process-
ing and analysis, but also hold considerable promise for
advancing the realms of medical personalization and pre-
cision. To fully utilize the value of perioperative big data,
a synergistic fusion of medical domain expertise and cut-
ting-edge data science technology is imperative. is col-
laborative approach can facilitate in-depth analysis and
exploration, furnishing a scientific and precise founda-
tion to inform clinical decision-making.
3 Introduction toperioperative big data
algorithms
Machine learning constitutes a methodology for attain-
ing artificial intelligence (AI) by enabling computational
systems to acquire patterns and rules from data, thereby
automating the processing and decision-making of tasks.
Categorically, fundamental machine learning algorithms
can be delineated into four primary classifications based
on the nature of the tasks they are designed to address:
supervised learning, unsupervised learning, semi-super-
vised learning, and reinforcement learning (RL) [9, 10].
Supervised machine learning aims to predict desired
outputs based on input data, operating on the princi-
ple of identifying input–output associations during the
training phase and using these associations to predict
the correct output for new cases. Common algorithms
in this category include support vector machines, logistic
regression, random forests, and others. Unsupervised
machine learning aims to infer potential patterns within
unlabeled data; the common algorithms include K-means
clustering, principal component analysis, and Gaussian
mixture model [11]. Semi-supervised learning uses both
labeled and unlabeled data to improve model perfor-
mance; the common algorithms include semi-supervised
clustering and semi-supervised classification. Reinforce-
ment learning learns through the interaction between
agents and the environment to maximize cumulative
rewards. Algorithms in reinforcement learning encom-
pass Q-learning, deep reinforcement learning, policy gra-
dients, and more [12].
In addition, deep learning as a powerful machine learn-
ing algorithm, has made remarkable strides in processing
high-dimensional complex data and addressing intricate
tasks by constructing multi-layer neural networks to
learn complex data feature representation. Deep learning
includes a spectrum of algorithms, such as convolutional
neural networks (CNNs) and recurrent neural networks,
and has already been technically applied in computer
vision, natural language processing, reinforcement learn-
ing, and generalized deep learning [1]. Large language
model, based on deep neural networks, has emerged as
a technology that learns language patterns and rules of
massive text data to comprehend, generate, and process
textual information.
In recent years, the large language model has made
important achievements in text generation, machine
translation, sentiment analysis and other tasks [13]. In
the healthcare domain, large language model can be
used to extract medical knowledge [14] and assist clini-
cal decision-making [15]. Considering the characteristics
of perioperative multi-objective optimization and multi-
decision parallel, in the future, based on the deep learn-
ing structure and strong representation ability of large
language model, it can conduct in-depth analysis of com-
plex medical texts and patient data to complete multi-
task intervention decision-making. For the requirements
of continuous dynamic prediction and decision-making,
reinforcement learning can be combined with multiple
trials and iterations in the simulation environment to find
the optimal decision-making strategy.
4 Applications ofbig data inperioperative
anesthesia management
e application of perioperative big data requires a series
of processes including data collection, data preprocess-
ing, data annotation, model construction, and model
evaluation (Fig.2). Due to the characteristics of periop-
erative big data, collected data must undergo integration
and cleaning processes. Subsequently, feature engineer-
ing is conducted, wherein specific indicators are selected

Page 4 of 11
Zhuetal. Anesthesiology and Perioperative Science (2024) 2:30
for model construction. Following model establishment,
validation and optimization procedures are undertaken
to enable clinical applicability. ese big data applications
are built upon comprehensive data collection and rigor-
ous analysis, serving as powerful tools to enhance vari-
ous aspects of perioperative care. is review will explore
the types of perioperative big data applications and how
they function across different scenarios and stages of the
perioperative period.
4.1 Perioperative risk prediction
Utilizing perioperative data such as demographic infor-
mation, vital signs and laboratory tests in conjunction
with machine learning algorithms, enables the develop-
ment of predictive models for early warning of adverse
events across different stages and scenarios. Given that
perioperative data integrates multiple sources and types
of data, it offers doctors a more comprehensive data per-
spective, allowing for the rapid and accurate identifica-
tion of high-risk patients. is enhances the efficiency,
quality, and safety of medical decision-making and surgi-
cal procedures.
Intraoperative management involves abundant surgi-
cal and anesthetic data, including physiological parame-
ters and drug usage among others. e application of big
data for predicting intraoperative adverse events enables
Fig.2 The process of big data application

Page 5 of 11
Zhuetal. Anesthesiology and Perioperative Science (2024) 2:30
personalized risk assessment based on patient-specific
characteristics. is facilitates real-time alerts to doctors,
providing effective support for the safety of surgery and
anesthesia. Recent studies have indicated that leverag-
ing perioperative big data and AI algorithms can achieve
early prediction. For the monitoring of depth of anesthe-
sia (DoA), various studies have now established multiple
machine learning models for the classification of DoA
[16–18]. In addition, corresponding predictive models
have been developed for intraoperative adverse events,
such as hypotension [19, 20], hypoxemia [21], bradycar-
dia [22], and massive transfusion events [23]. ese will
allow anesthesiologists to recognize risks at an early stage
and take measures to prevent their occurrence. Of note,
the best area under the receiver operating characteristic
curve (AUROC) in the external validated dataset among
these models can reach 0.943. In clinical practice, a high
AUROC value implies that the model can predict events
more accurately [24], thereby assisting physicians in mak-
ing timely interventions.
Postoperative complications have the potential to
impede patient recovery, prolong the length of hospital
stay, and increase mortality [25, 26]. Early identification
of high-risk patients plays a pivotal role in enabling clini-
cians to establish accurate diagnoses and treatment strat-
egies. In the domain of big data research, various studies
have yielded pertinent findings for diverse surgical proce-
dures and types of complications. In the fields of postop-
erative respiratory failure [27], acute kidney injury [28],
major adverse cardiac and cerebrovascular events [29],
and postoperative delirium [30], researchers have already
established predictive models for these complications.
In addition to accomplishing static predictions, some
of these models can achieve dynamic forecasting [28],
thereby meeting the requirements for real-time monitor-
ing and assessment of patient condition during the perio-
perative period.
Regarding perioperative anesthesia management, the
majority of current research focuses on warning studies.
However, predictive models built on a single dataset are
gradually facing bottlenecks, including issues related to
data quality, model robustness and algorithms of periop-
erative big data. Given the characteristics of perioperative
big data, there is an urgent need for the use of multicenter
data for model construction and the implementation of
data quality enhancement techniques. For cross-hospital
multi-center databases, establishing standardized medi-
cal terminologies and data structures is crucial. Moreo-
ver, with the support of existing public databases such
as the Medical Information Mart for Intensive Care
(MIMIC) dataset [31] and eICU Collaborative Research
Database [32], external validation can be achieved to
address deficiencies in the current multi-center datasets.
Meanwhile, to address algorithm bottlenecks, innovation
and optimization of algorithm strategies are necessary to
enhance model performance.
4.2 Perioperative anesthesia decision‑making
Making accurate medical decisions is challenging, par-
ticularly within the dynamic and continuous context of
perioperative anesthesia management. is complexity
significantly escalates in decision-making studies com-
pared to predictive research. Consequently, there is an
imperative shift towards leveraging big data for research
to assist and enhance decision-making processes in this
critical domain. In the complexity of perioperative set-
tings, where patients may encounter adverse events at
any moment, perioperative decision models can recog-
nize and provide decision support in a timely manner,
ensuring optimal care pathways are promptly enacted.
4.2.1 Optimization ofanesthetic decisions inenhanced
recovery aftersurgery (ERAS)
Perioperative multi-objective optimization, which is an
optimization method that considers multiple compet-
ing goals or indicators in the decision-making problem,
needs to be implemented. is involves balancing mul-
tiple interrelated and sometimes conflicting objectives
during surgery to ensure patient safety. In the periopera-
tive period, multiple key factors need to be balanced to
ensure surgical success and patient recovery, such as the
dose of anesthetic drugs, fluid management, blood pres-
sure control, oxygenation status and antibiotic use [33].
ERAS represents an approach to perioperative care that
involves the implementation of multi-objective opti-
mization strategies. ERAS, built upon evidence-based
medicine, constitutes a strategic approach targeting sur-
gical and psychological stress to reduce complications,
decrease the duration of postoperative hospitalization
and lower risks of readmission and mortality [6, 7]. Given
the interrelated nature of ERAS measures spanning mul-
tiple departments, challenges persist in optimizing ERAS
interventions and enhancing decision-making in periop-
erative management.
Bayesian network (BN), originating from statistics,
computer science, and artificial intelligence was initially
proposed by Pearl [34]. BN [35, 36], a form of machine
learning, serves as a graphical model for describing rela-
tionships between variables. BN is particularly adept at
representing the joint probability distribution in a con-
cise and intuitive manner. Moreover, BN stands as one
of the most effective theoretical models in the realms of
uncertain knowledge and probability reasoning. Applying
BN to perioperative ERAS management holds the prom-
ise of developing a clinical decision support system tai-
lored for anesthesia perioperative management, thereby

Page 6 of 11
Zhuetal. Anesthesiology and Perioperative Science (2024) 2:30
supporting ERAS teams in devising personalized reha-
bilitation plans for patients. In the domain of medical
decision-making, BN has already demonstrated its utility.
Hargrave etal. [37]. introduced a BN and complemen-
tary visualization tool designed for decision support in
online cone-beam computed tomography (CBCT)-based
image-guided radiotherapy for prostate cancer patients.
e study utilized relevant indicators in differences of
CBCT images as nodes to construct a BN, aiming to pro-
vide patients with recommendations regarding continued
treatment, similar to various decisions needed during the
perioperative period. In the future, BN for clinical prac-
tice could be established to optimize decision manage-
ment. Huang etal. [38] established and validated a BN
model using large-scale influenza post-marketing vaccine
surveillance data, identifying age, gender, and specific
adverse events as significant factors for the prediction
of Guillain-Barré syndrome risk. e BN model effec-
tively captures complex interactions among post-vacci-
nation symptoms and provides accurate risk estimates.
In perioperative management, this type of BN model can
also be used to screen high-risk factors and explore the
relationships between these risk factors. Based on the
characteristics of BN algorithms, our team conducted
an application attempt of BN in the ERAS field, and
achieved decision recommendations in anesthetic deci-
sions for gynecological surgeries [39]. We developed a
method for optimizing anesthetic decisions in ERAS and
then investigated the relationship between these anes-
thetic decisions and outcomes. Subsequently, pivotal
decisions were identified and composite recommenda-
tions for these decisions were provided. erefore, BN
can be employed to establish a network of relationships
between decisions and outcomes, elucidating connec-
tions among indicators. By utilizing conditional prob-
ability tables, key decisions can be identified, and optimal
combination strategies can be explored, thereby achiev-
ing the goal of multi-objective optimization.
4.2.2 Anesthetic decision‑making incritically ill patients
For the anesthesia of critically ill patients, the adoption of
advanced decision-making techniques has become a piv-
otal point in this field. Markov decision processes (MDPs)
stand out as a structured and systematic approach within
this context. MDPs provide a mathematical framework
for modeling sequential decision-making under uncer-
tainty [40]. In the intricate landscape of critical care,
where patient conditions are dynamic and uncertain,
MDPs offer an invaluable tool for doctors. e utilization
of MDPs in anesthesia decision-making allows for con-
sideration of the stochastic nature of patient responses
and the unpredictable evolution of medical condi-
tions. is comprehensive approach enables clinicians
to optimize decisions based on the evolving state of the
patient’s health. Whether it involves adjusting anesthetic
dosages, ventilation parameters, or anticipating potential
complications, MDPs facilitate a nuanced decision-mak-
ing strategy. By incorporating probabilistic transitions
between different states, MDPs offer a forward-looking
perspective, allowing for anticipatory interventions tai-
lored to individual patient needs.
Complementing the role of MDPs, RL introduces a
dynamic and adaptive dimension to anesthesia decision-
making for critically ill patients. RL, a subset of machine
learning, focuses on training agents to make sequential
decisions through interactions with the environment. In
RL, the interaction between the agent’s task and the envi-
ronment can be modeled as an MDP. MDP provides a
structured framework for describing state transitions and
rewards, and reinforcement learning algorithms leverage
this information to learn the optimal strategy. In critical
care scenarios characterized by rapid changes, RL algo-
rithms continuously refine their strategies based on real-
time feedback. For instance, in the context of sedative
administration, Padmanabhan etal. [41] proposed a RL
based optimal adaptive control approach for the continu-
ous infusion of a sedative to maintain a required level of
sedation. For the use of vasoactive drugs and fluid man-
agement, some research teams have conducted research
in the field of sepsis. Komorowski et al. [42] used the
MIMIC-III to develop a computational model based on
reinforcement learning and to test it through the eICU
Research Institute Database. is model can dynamically
suggest optimal treatments for adult patients with sep-
sis in the intensive care unit. e results of this research
support the strategy of early use of low-dose vasopres-
sor in the management of sepsis, potentially avoiding
administration of excessive amounts of fluids. Further-
more, the model facilitates personalized treatment plan-
ning for each patient, thereby contributing to enhanced
patient outcomes. In the context of mechanical ventila-
tion and extubation [43, 44], some related studies have
also employed modeling and used reinforcement learning
to find strategies.
4.2.3 Closed‑loop anesthetic management
Closed-loop anesthetic management represents a trans-
formative approach in the administration of anesthesia,
characterized by real-time monitoring and automated
adjustments. Closed-loop systems leverage continuous
feedback from monitoring devices to dynamically regu-
late anesthetic drug dosages, ensuring that the depth of
anesthesia aligns with predefined parameters. Existing
closed-loop systems include both single and multi-system
approaches. ese encompass the regulation of propo-
fol using the bispectral index [45–51] and the control of

Page 7 of 11
Zhuetal. Anesthesiology and Perioperative Science (2024) 2:30
propofol, remifentanil, and rocuronium from the per-
spectives of sedation, analgesia, and muscle relaxation
[52–54]. Additionally, closed-loop systems involving
fluid therapy [55–58] and vasopressor [59–63] are also
present. However, these closed-loop systems exhibit rela-
tively singular input parameters, lacking a more extensive
array of input parameters, which makes them unable to
implement the complete anesthesia process. Additionally,
conventional anesthesia delivery systems are typically
based on fixed mathematical formulae and algorithms.
ese systems utilize physiological models and predeter-
mined rules to regulate the administration of anesthetics.
While such systems may offer high reliability and predict-
ability due to their reliance on established mathematical
principles, they are often less flexible compared to AI-
based systems. AI-based systems can leverage advanced
algorithms and machine learning techniques to adapt and
optimize anesthesia delivery in real-time, offering greater
flexibility in response to patient-specific factors and
dynamic surgical conditions.
Recent advancements in closed-loop anesthetic
management systems focus on refining algorithms,
enhancing adaptability to individual patient needs,
and incorporating additional monitoring parameters.
For instance, Ren etal. [64] established an "AI anesthe-
siologist" model tailored for intraoperative dosing of
personalized anesthetic drugs (propofol and remifen-
tanil) based on a convolutional neural network, com-
bined with both sliding window sampling method and
residual learning module. e results revealed that
the medication distribution predicted by the model
could be equivalent to the actual medication distribu-
tion provided by the clinical anesthesiologist. With
the assistance of AI algorithms and the integration of
multidimensional information, including patients’ vital
signs, comorbidities, medications, and contextual inter-
actions, the future development of anesthesia robots is
anticipated. In addition to drug regulation, considera-
tion must be given to various perioperative scenarios.
Furthermore, the integration of expert knowledge
during model construction is essential. Only then can
anesthesia robots assist anesthesiologists in decision-
making, facilitating the accomplishment of true closed-
loop anesthesia management.
4.3 Perioperative auxiliary examination
4.3.1 Ultrasound positioning
Ultrasound imaging serves as a noninvasive modality
for the visualization of anatomical structures, includ-
ing nerves, musculoskeletal structures, needles used and
anesthetics injected. Ultrasound images have the charac-
teristics of real-time, dynamic, large and complex data.
AI technology has been harnessed to establish models
rooted in source data, thereby assisting doctors in the
identification and localization of pertinent anatomi-
cal structures. is application of AI holds the potential
to reduce complications, such as hematoma formation
and nerve injuries. For regional nerve blocks, Berggreen
et al. [65] proposed an approach using a deep learn-
ing semantic segmentation model with U-net architec-
ture to identify the femoral nerve in ultrasound images.
Concurrently, specialized AI-based devices, like Scan-
Nav Anatomy Peripheral Nerve Block (Intelligent Ultra-
sound, Cardiff, UK), have been developed [66]. Bowness
etal. [67] have evaluated the accuracy of the artificial-
intelligence color overlay and its perceived influence on
the risk of adverse events or block failure. e artificial
intelligence models identified the structure of inter-
est in 93.5% of cases (1519/1624), with a false-negative
rate of 3.0% (48/1624) and a false-positive rate of 3.5%
(57/1624). In terms of spinal anesthesia, Chan etal. [68]
developed an ultrasound-guided automated spinal land-
mark identification program to assist anesthetists on spi-
nal needle insertion point with a graphical user interface
for spinal anesthesia. ere is a good correlation between
program recorded and clinician measured depth of the
skin to posterior complex of dura mater. Pesteie et al.
[69] proposed a deep network architecture along with a
feature augmentation technique, designed for automatic
identification of the anatomical landmarks of the epidural
space in ultrasound images. is could help automatically
localize the needle target for epidural needle placement.
4.3.2 Airway management
Airway management is critical in ensuring the safety and
quality of anesthesia. An unpredicted difficult airway
can result in failed intubation and, if ventilation is diffi-
cult, can evolve to a "cannot ventilate, cannot oxygenate"
scenario, with risk of hypoxic-ischemic brain injury, and
death [70]. erefore, it is important to identify poten-
tially difficult airways. Tavolara etal. [71] proposed an
ensemble of convolutional neural networks which lever-
ages a database of celebrity facial images to learn robust
features of multiple face regions. And a deep learning
model was constructed to identify difficult-to-intubate
patients using frontal face images, which both exceeds
the sensitivity and specificities of conventional bedside
tests as well as common deep learning methods. As for
difficult mask ventilation, Pei etal. [72] developed a pre-
diction model with morphometric data and machine
learning (ML) algorithms, while the logistic regression
model performed best among the 10 ML models achiev-
ing an AUROC of 0.825.
Big data analysis has also been applied in the process of
intubation. Matava etal. [73] devised a model capable of

Page 8 of 11
Zhuetal. Anesthesiology and Perioperative Science (2024) 2:30
real-time classification of vocal cords and tracheal airway
anatomy during video laryngoscopy or bronchoscopy,
leveraging three convolutional neural networks. e
overall confidence of classification for the vocal cords
and tracheal rings for ResNet, Inception and MobileNet
CNNs were as follows: 0.84, 0.78, and 0.64 for vocal
cords, respectively, and 0.69, 0.72, 0.54 for tracheal rings,
respectively.
5 Future perspectives
e prospects for big data in perioperative management
are promising, encompassing all phases from preopera-
tive to intraoperative and postoperative periods, which
can bring benefits from multiple perspectives (Fig. 3).
It offers the capability to predict perioperative risks and
provides valuable support during perioperative proce-
dures. By leveraging big data analytics, doctors can attain
a deeper comprehension of patient outcomes, thereby
facilitating the formulation of more efficacious treat-
ment strategies. is, in turn, leads to an enhancement
in the overall quality of healthcare delivery to patients.
Furthermore, the integration of big data has the capacity
to yield cost reductions and foster improved communi-
cation between doctors and their patients, thereby aug-
menting the overall patient experience. In the context of
the evolving landscape of big data research, the field of
perioperative management has gained a fresh perspec-
tive and novel methodologies. As a result, the discipline
of anesthesiology is poised to achieve breakthrough
advancements.
Currently, the majority of AI research in the fields of
anesthesiology and critical care medicine remain at the
stage of retrospective analysis or prototype development,
with only a few applications successfully integrated into
clinical practice. ere are several reasons for this. Firstly,
the lack of multi-center and high-quality datasets severely
constrains the robustness of models. Secondly, the appli-
cation of big data research requires interdisciplinary
collaboration, necessitating the integration of expertise
from various fields, overcoming communication barriers,
and standardizing terminology and methodologies across
different domains, which poses challenges. Additionally,
clinicians require transparent and easily understand-
able models to trust and incorporate AI-driven insights
into patient care decisions, yet most AI algorithms lack
interpretability. Data security and privacy concerns are
also significant challenges, demanding stringent protec-
tive measures to safeguard patient information. Ethical
considerations, including bias, responsibility and unin-
tended consequences, further complicate the process
of integrating AI into clinical practice. Finally, the lack
of standardized processes for utilizing AI in healthcare
has hindered the translation of AI research findings into
routine patient care. Addressing these challenges will be
critical in fully harnessing the potential of big data and AI
technology to enhance perioperative anesthesia manage-
ment and critical care medicine.
"Garbage in, garbage out" underscores the fundamen-
tal truth that high-quality data is indispensable for the
successful deployment of big data and machine learning
models. To ensure stability and robustness, the model
needs to identify the authentic underlying relationships
within the data. However, noise inherent in low-quality
data can disrupt these connections. Perioperative big
data, characterized by high dimensionality, heterogene-
ous sources, and high noise, necessitates the use of data
governance techniques such as outlier detection, miss-
ing value imputation, feature selection, and dimensional-
ity reduction to enhance data quality, thereby improving
the accuracy and robustness of models. Furthermore,
conducting multicenter collaborative research to enrich
data diversity and volume has also become one of the
methods to improve model accuracy and robustness. In
addition to cross-center data issues, there is also a need
to address interdisciplinary collaboration. To tackle the
challenge of integrating expertise from different domains,
entity extraction and relationship recognition techniques
Fig.3 Benefits from the application of big data in the perioperative period

Page 9 of 11
Zhuetal. Anesthesiology and Perioperative Science (2024) 2:30
can be employed to extract professional knowledge from
literature, expertise, and other data sources for the con-
struction of a cross-domain knowledge graph. When
dealing with specific perioperative scenarios, traditional
methods may not fully meet the requirements, necessi-
tating algorithmic innovation. For example, optimizing
perioperative drug therapy involves numerous dosage
combinations that traditional machine learning methods
struggle with, making reinforcement learning or deep
reinforcement learning a superior alternative [74, 75].
Furthermore, in the aspect of data security, the imple-
mentation of data encryption technologies is essential.
is includes algorithms such as Message Digest Algo-
rithm5 [76], as well as access control and anonymization
techniques [77]. Additionally, the adoption of distributed
machine learning technologies, such as federated learn-
ing, is vital. is approach enables local data processing,
thereby ensuring the secure and efficient deployment
of big data technologies. Achieving robust data security
is crucial for the widespread adoption of perioperative
applications.
Multi-objective optimization faces challenges such as
objective conflicts, complexity in non-dominated solu-
tion sets, high dimensionality, computational costs, con-
vergence of solutions, problem representation, and weight
settings. Most of the existing models rely on data within
specific time frames or event scope for modeling and lack
dynamic data, yet dynamic monitoring, real-time pre-
diction, and intervention recommendation are clinical
requirements. Future advancements will involve leverag-
ing technologies like reinforcement learning and large
language model to mine big data, continuously enhance
application functionalities, expand perioperative process
tasks, and further develop artificial intelligence applications
or intelligent robots for doctors.
Abbreviations
AI Artificial intelligence
AUROC Area under the receiver operating characteristic curve
BN Bayesian network
CBCT Cone-beam computed tomography
CNNs Convolutional neural networks
DoA Depth of anesthesia
ERAS Enhanced recovery after surgery
MDPs Markov decision processes
MIMIC Medical information mart for intensive care
ML Machine learning
RL Reinforcement learning
Authors’ contributions
Conceptualization: Bin Yi; Literature search: Yiziting Zhu, Xiang Liu; Literature
summary: Yiziting Zhu; Writing—original draft preparation: Yiziting Zhu, Xiang
Liu, Yujie Li; Writing—review and editing: Yujie Li, Bin Yi; Funding acquisition:
Yujie Li, Bin Yi.
Funding
This study was funded by National Key R&D Program of China (No.
2018YFC0116702), National Science Foundation of China (No. 82100658, No.
82070630), Chongqing Talents Project (No. CQYC202103080), Chongqing
Talents Project (No. 4139Z2396), special support for Chongqing postdoctoral
research project in 2020 (missing the grant number?) and Natural Science
Foundation of Chongqing (Key Project) (No. CSTB2023NSCQ-ZDJ0005).
Availability of data and materials
Not applicable.
Declarations
Ethics approval and consent to participate
Not applicable.
Competing interests
On behalf of all authors, the corresponding author states that there is no
conflict of interest.
Received: 29 December 2023 Revised: 1 August 2024 Accepted: 4 August
2024
References
1. Esteva A, Robicquet A, Ramsundar B, Kuleshov V, DePristo M, Chou K, et al.
A guide to deep learning in healthcare. Nat Med. 2019;25(1):24–9. https://
doi. org/ 10. 1038/ s41591- 018- 0316-z.
2. Mahmoudian M, Zanjani SM, Shahinzadeh H, Kabalci Y, Kabalci E,
Ebrahimi F, editors. An overview of big data concepts, methods, and
analytics: challenges, issues, and opportunities. Proceedings of the 5th
global power, energy and communication conference; 2023 Jun 14–16;
Cappadocia, Turkiye. New York: IEEE; 2023.
3. Buch VH, Ahmed I, Maruthappu M. Artificial intelligence in medicine:
current trends and future possibilities. Br J Gen Pract. 2018;68(668):143–4.
https:// doi. org/ 10. 3399/ bjgp1 8X695 213.
4. Churpek MM, Yuen TC, Winslow C, Meltzer DO, Kattan MW, Edelson DP.
Multicenter comparison of machine learning methods and conven-
tional regression for predicting clinical deterioration on the wards. Crit
Care Med. 2016;44(2):368–74. https:// doi. org/ 10. 1097/ ccm. 00000 00000
001571.
5. Lee HC, Yoon SB, Yang SM, Kim WH, Ryu HG, Jung CW, et al. Prediction
of acute kidney injury after liver transplantation: machine learning
approaches vs. logistic regression model. J Clin Med. 2018;7(11):428.
https:// doi. org/ 10. 3390/ jcm71 10428.
6. Gustafsson UO, Oppelstrup H, Thorell A, Nygren J, Ljungqvist O. Adher-
ence to the ERAS protocol is associated with 5-year survival after
colorectal cancer surgery: a retrospective cohort study. World J Surg.
2016;40(7):1741–7. https:// doi. org/ 10. 1007/ s00268- 016- 3460-y.
7. Ripollés-Melchor J, Ramírez-Rodríguez JM, Casans-Francés R, Aldecoa
C, Abad-Motos A, Logroño-Egea M, et al. Association between use of
enhanced recovery after surgery protocol and postoperative com-
plications in colorectal surgery: The Postoperative Outcomes Within
Enhanced Recovery After Surgery Protocol (POWER) study. JAMA Surg.
2019;154(8):725–36. https:// doi. org/ 10. 1001/ jamas urg. 2019. 0995.
8. Mohan A, Blough DM, Kurc T, Post A, Saltz J, editors. Detection of conflicts
and inconsistencies in taxonomy-based authorization policies. Proceed-
ings of the 2011 IEEE International Conference on Bioinformatics and
Biomedicine; 2011 Nov 12–15; Atlanta, USA. New York: IEEE; 2012. https://
doi. org/ 10. 1109/ bibm. 2011. 79.
9. Deo RC. Machine learning in medicine. Circulation. 2015;132(20):1920–30.
https:// doi. org/ 10. 1161/ circu latio naha. 115. 001593.
10. Russell SJ, Norvig P. Artificial intelligence: a modern approach. 2nd ed.
Hoboken: Prentice Hall; 2002.
11. Yu KH, Beam AL, Kohane IS. Artificial intelligence in healthcare.
Nat Biomed Eng. 2018;2(10):719–31. https:// doi. org/ 10. 1038/
s41551- 018- 0305-z.
12. Sutton RS, Barto AG. Reinforcement learning: an Introduction. IEEE Trans-
act Neur Netw. 1998;9(5):1054. https:// doi. org/ 10. 1109/ TNN. 1998. 712192.
13. Samant RM, Bachute MR, Gite S, Kotecha K. Framework for deep learning-
based language models using multi-task learning in natural language

Page 10 of 11
Zhuetal. Anesthesiology and Perioperative Science (2024) 2:30
understanding: a systematic literature review and future directions. IEEE
Access. 2022;10:17078–97. https:// doi. org/ 10. 1109/ ACCESS. 2022. 31497
98.
14. Li H, Wu Y, Schlegel V, Batista-Navarro R, Nguyen T-T, Kashyap AR, et al.
PULSAR: Pre-training with extracted healthcare terms for summaris-
ing patients’ problems and data augmentation with black-box large
language models. arXiv:2306.02754 [Preprint]. 2023. [cited 2023 Dec 25]:
[7 p.]. Available from: https:// doi. org/ 10. 48550/ arXiv. 2306. 02754.
15. Yeo YH, Samaan JS, Ng WH, Ting PS, Trivedi H, Vipani A, et al. Assessing the
performance of ChatGPT in answering questions regarding cirrhosis and
hepatocellular carcinoma. Clin Mol Hepatol. 2023;29(3):721–32. https://
doi. org/ 10. 3350/ cmh. 2023. 0089.
16. Saadeh W, Khan FH, Altaf MAB. Design and implementation of a machine
learning based EEG processor for accurate estimation of depth of anes-
thesia. IEEE Trans Biomed Circuits Syst. 2019;13(4):658–69. https:// doi. org/
10. 1109/ tbcas. 2019. 29218 75.
17. Park Y, Han SH, Byun W, Kim JH, Lee HC, Kim SJ. A real-time depth of
anesthesia monitoring system based on deep neural network with large
EDO tolerant EEG analog front-end. IEEE Trans Biomed Circuits Syst.
2020;14(4):825–37. https:// doi. org/ 10. 1109/ tbcas. 2020. 29981 72.
18. Dubost C, Humbert P, Oudre L, Labourdette C, Vayatis N, Vidal PP. Quan-
titative assessment of consciousness during anesthesia without EEG
data. J Clin Monit Comput. 2021;35(5):993–1005. https:// doi. org/ 10. 1007/
s10877- 020- 00553-4.
19. Hwang E, Park YS, Kim JY, Park SH, Kim J, Kim SH. Intraoperative hypoten-
sion prediction based on features automatically generated within an
interpretable deep learning model. IEEE Trans Neural Netw Learn Syst.
2023:1–15. https:// doi. org/ 10. 1109/ tnnls. 2023. 32731 87.
20. Lee S, Lee HC, Chu YS, Song SW, Ahn GJ, Lee H, et al. Deep learning
models for the prediction of intraoperative hypotension. Br J Anaesth.
2021;126(4):808–17. https:// doi. org/ 10. 1016/j. bja. 2020. 12. 035.
21. Lundberg SM, Nair B, Vavilala MS, Horibe M, Eisses MJ, Adams T, et al.
Explainable machine-learning predictions for the prevention of hypoxae-
mia during surgery. Nat Biomed Eng. 2018;2(10):749–60. https:// doi. org/
10. 1038/ s41551- 018- 0304-0.
22. Solomon SC, Saxena RC, Neradilek MB, Hau V, Fong CT, Lang JD, et al.
Forecasting a crisis: machine-learning models predict occurrence of
intraoperative bradycardia associated with hypotension. Anesth Analg.
2020;130(5):1201–10. https:// doi. org/ 10. 1213/ ane. 00000 00000 004636.
23. Lee SM, Lee G, Kim TK, Le T, Hao J, Jung YM, et al. Development and
validation of a prediction model for need for massive transfusion during
surgery using intraoperative hemodynamic monitoring data. JAMA Netw
Open. 2022;5(12):e2246637. https:// doi. org/ 10. 1001/ jaman etwor kopen.
2022. 46637.
24. McDermott M, Hansen LH, Zhang H, Angelotti G, Gallifant J. A closer look
at AUROC and AUPRC under class imbalance. arXiv:2401.06091 [Preprint].
2024. p. 32. [cited 2023 Dec 26]. Available from: https:// doi. org/ 10. 48550/
arXiv. 2401. 06091.
25. Devereaux PJ, Chan MT, Alonso-Coello P, Walsh M, Berwanger O, Villar
JC, et al. Association between postoperative troponin levels and 30-day
mortality among patients undergoing noncardiac surgery. JAMA.
2012;307(21):2295–304. https:// doi. org/ 10. 1001/ jama. 2012. 5502.
26. Devereaux PJ, Sessler DI. Cardiac complications in patients undergoing
major noncardiac surgery. N Engl J Med. 2015;373(23):2258–69. https://
doi. org/ 10. 1056/ NEJMr a1502 824.
27. Kiyatkin ME, Aasman B, Fazzari MJ, Rudolph MI, Vidal Melo MF, Eikermann
M, et al. Development of an automated, general-purpose prediction tool
for postoperative respiratory failure using machine learning: A retrospec-
tive cohort study. J Clin Anesth. 2023;90:111194. https:// doi. org/ 10. 1016/j.
jclin ane. 2023. 111194.
28. Ryan CT, Zeng Z, Chatterjee S, Wall MJ, Moon MM, Coselli JS, et al.
Machine learning for dynamic and early prediction of acute kidney injury
after cardiac surgery. J Thorac Cardiovasc Surg. 2022. https:// doi. org/ 10.
1016/j. jtcvs. 2022. 09. 045.
29. Mahajan A, Esper S, Oo TH, McKibben J, Garver M, Artman J, et al. Devel-
opment and validation of a machine learning model to identify patients
before surgery at high risk for postoperative adverse events. JAMA Netw
Open. 2023;6(7):e2322285. https:// doi. org/ 10. 1001/ jaman etwor kopen.
2023. 22285.
30. Dodsworth BT, Reeve K, Falco L, Hueting T, Sadeghirad B, Mbuagbaw L,
et al. Development and validation of an international preoperative risk
assessment model for postoperative delirium. Age Ageing. 2023;52(6).
https:// doi. org/ 10. 1093/ ageing/ afad0 86.
31. Johnson AEW, Bulgarelli L, Shen L, Gayles A, Shammout A, Horng S, et al.
MIMIC-IV, a freely accessible electronic health record dataset. Sci Data.
2023;10(1):1. https:// doi. org/ 10. 1038/ s41597- 022- 01899-x.
32. Pollard TJ, Johnson AEW, Raffa JD, Celi LA, Mark RG, Badawi O. The eICU
collaborative research database, a freely available multi-center database
for critical care research. Sci Data. 2018;5:180178. https:// doi. org/ 10. 1038/
sdata. 2018. 178.
33. Brennan M, Hagan JD, Giordano C, Loftus TJ, Price CE, Aytug H, et al. Multi-
objective optimization challenges in perioperative anesthesia: a review.
Surgery. 2021;170(1):320–4. https:// doi. org/ 10. 1016/j. surg. 2020. 11. 005.
34. Pearl J. Probabilistic reasoning in intelligent systems: networks of plausi-
ble inference. San Francisco: Morgan Kaufmann; 1988.
35. Koller D, Friedman N. Probabilistic graphical models: principles and tech-
niques. 1st ed. Cambridge: MIT Press; 2009.
36. Kyburg HE. Probabilistic reasoning in intelligent systems. J Philos.
1991;88(8):434–7.
37. Hargrave C, Deegan T, Bednarz T, Poulsen M, Harden F, Mengersen K. An
image-guided radiotherapy decision support framework incorporating
a Bayesian network and visualization tool. Med Phys. 2018;45(7):2884–97.
https:// doi. org/ 10. 1002/ mp. 12979.
38. Huang Y, Luo C, Jiang Y, Du J, Tao C, Chen Y, et al. A Bayesian network
to predict the risk of post influenza vaccination Guillain-Barré syn-
drome: development and validation study. JMIR Public Health Surveill.
2022;8(3):e25658. https:// doi. org/ 10. 2196/ 25658.
39. Chen Y, Zhu Y, Zhong K, Yang Z, Li Y, Shu X, et al. Optimization of
anesthetic decision-making in ERAS using Bayesian network. Front Med
(Lausanne). 2022;9:1005901. https:// doi. org/ 10. 3389/ fmed. 2022. 10059 01.
40. Alagoz O, Hsu H, Schaefer AJ, Roberts MS. Markov decision processes: a
tool for sequential decision making under uncertainty. Med Decis Mak-
ing. 2010;30(4):474–83. https:// doi. org/ 10. 1177/ 02729 89x09 353194.
41. Padmanabhan R, Meskin N, Haddad WM. Optimal adaptive control
of drug dosing using integral reinforcement learning. Math Biosci.
2019;309:131–42. https:// doi. org/ 10. 1016/j. mbs. 2019. 01. 012.
42. Komorowski M, Celi LA, Badawi O, Gordon AC, Faisal AA. The artificial
intelligence clinician learns optimal treatment strategies for sepsis in
intensive care. Nat Med. 2018;24(11):1716–20. https:// doi. org/ 10. 1038/
s41591- 018- 0213-5.
43. Yu C, Liu J, Zhao H. Inverse reinforcement learning for intelligent
mechanical ventilation and sedative dosing in intensive care units. BMC
Med Inform Decis Mak. 2019;19(Suppl 2):57. https:// doi. org/ 10. 1186/
s12911- 019- 0763-6.
44. Yu C, Ren G, Dong Y. Supervised-actor-critic reinforcement learning for
intelligent mechanical ventilation and sedative dosing in intensive care
units. BMC Med Inform Decis Mak. 2020;20(Suppl 3):124. https:// doi. org/
10. 1186/ s12911- 020- 1120-5.
45. Struys MM, De Smet T, Versichelen LF, Van De Velde S, Van den Broecke
R, Mortier EP. Comparison of closed-loop controlled administration of
propofol using Bispectral Index as the controlled variable versus "stand-
ard practice" controlled administration. Anesthesiology. 2001;95(1):6–17.
https:// doi. org/ 10. 1097/ 00000 542- 20010 7000- 00007.
46. De Smet T, Struys MM, Greenwald S, Mortier EP, Shafer SL. Estimation of
optimal modeling weights for a Bayesian-based closed-loop system for
propofol administration using the bispectral index as a controlled vari-
able: a simulation study. Anesth Analg. 2007;105(6):1629–38. https:// doi.
org/ 10. 1213/ 01. ane. 00002 87269. 06170. 0f.
47. Liu N, Chazot T, Genty A, Landais A, Restoux A, McGee K, et al. Titration
of propofol for anesthetic induction and maintenance guided by the
bispectral index: closed-loop versus manual control: a prospective, rand-
omized, multicenter study. Anesthesiology. 2006;104(4):686–95. https://
doi. org/ 10. 1097/ 00000 542- 20060 4000- 00012.
48. Puri GD, Kumar B, Aveek J. Closed-loop anaesthesia delivery system
(CLADS) using bispectral index: a performance assessment study.
Anaesth Intensive Care. 2007;35(3):357–62. https:// doi. org/ 10. 1177/ 03100
57x07 03500 306.
49. Hemmerling TM, Charabati S, Zaouter C, Minardi C, Mathieu PA. A
randomized controlled trial demonstrates that a novel closed-loop
propofol system performs better hypnosis control than manual admin-
istration. Can J Anaesth. 2010;57(8):725–35. https:// doi. org/ 10. 1007/
s12630- 010- 9335-z.

Page 11 of 11
Zhuetal. Anesthesiology and Perioperative Science (2024) 2:30
50. Moore BL, Doufas AG, Pyeatt LD. Reinforcement learning: a novel
method for optimal control of propofol-induced hypnosis. Anesth Analg.
2011;112(2):360–7. https:// doi. org/ 10. 1213/ ANE. 0b013 e3182 0334a7.
51. Neckebroek M, Boldingh JHL, De Smet T, Struys M. Influence of remifen-
tanil on the control performance of the bispectral index controlled
Bayesian-based closed-loop system for propofol administration. Anesth
Analg. 2020;130(6):1661–9. https:// doi. org/ 10. 1213/ ane. 00000 00000
004208.
52. Hemmerling TM, Arbeid E, Wehbe M, Cyr S, Taddei R, Zaouter C. Evalua-
tion of a novel closed-loop total intravenous anaesthesia drug delivery
system: a randomized controlled trial. Br J Anaesth. 2013;110(6):1031–9.
https:// doi. org/ 10. 1093/ bja/ aet001.
53. Wehbe M, Arbeid E, Cyr S, Mathieu PA, Taddei R, Morse J, et al. A technical
description of a novel pharmacological anesthesia robot. J Clin Monit
Comput. 2014;28(1):27–34. https:// doi. org/ 10. 1007/ s10877- 013- 9451-8.
54. Nagata O, Matsuki Y, Matsuda S, Hazama K, Fukunaga S, Nakatsuka
H, et al. Anesthesia management via an automated control system
for propofol, remifentanil, and rocuronium compared to manage-
ment by anesthesiologists: an investigator-initiated study. J Clin Med.
2023;12(20):6611. https:// doi. org/ 10. 3390/ jcm12 206611.
55. Joosten A, Coeckelenbergh S, Delaporte A, Ickx B, Closset J, Roumeguere
T, et al. Implementation of closed-loop-assisted intra-operative goal-
directed fluid therapy during major abdominal surgery: A case-control
study with propensity matching. Eur J Anaesthesiol. 2018;35(9):650–8.
https:// doi. org/ 10. 1097/ eja. 00000 00000 000827.
56. Rinehart J, Lilot M, Lee C, Joosten A, Huynh T, Canales C, et al. Closed-
loop assisted versus manual goal-directed fluid therapy during high-risk
abdominal surgery: a case-control study with propensity matching. Crit
Care. 2015;19(1):94. https:// doi. org/ 10. 1186/ s13054- 015- 0827-7.
57. Joosten A, Huynh T, Suehiro K, Canales C, Cannesson M, Rinehart J. Goal-
directed fluid therapy with closed-loop assistance during moderate risk
surgery using noninvasive cardiac output monitoring: a pilot study. Br J
Anaesth. 2015;114(6):886–92. https:// doi. org/ 10. 1093/ bja/ aev002.
58. Jin X, Bighamian R, Hahn JO. Development and in silico evaluation of a
model-based closed-loop fluid resuscitation control algorithm. IEEE Trans
Biomed Eng. 2018;66(7):1905–14. https:// doi. org/ 10. 1109/ tbme. 2018.
28809 27.
59. Ngan Kee WD, Khaw KS, Tam YH, Ng FF, Lee SW. Performance of a
closed-loop feedback computer-controlled infusion system for maintain-
ing blood pressure during spinal anaesthesia for caesarean section: a
randomized controlled comparison of norepinephrine versus phenyle-
phrine. J Clin Monit Comput. 2017;31(3):617–23. https:// doi. org/ 10. 1007/
s10877- 016- 9883-z.
60. Ngan Kee WD, Tam YH, Khaw KS, Ng FF, Lee SWY. Closed-loop feed-
back computer-controlled phenylephrine for maintenance of blood
pressure during spinal anesthesia for cesarean delivery: a randomized
trial comparing automated boluses versus infusion. Anesth Analg.
2017;125(1):117–23. https:// doi. org/ 10. 1213/ ane. 00000 00000 001974.
61. Joosten A, Delaporte A, Alexander B, Su F, Creteur J, Vincent JL, et al.
Automated titration of vasopressor infusion using a closed-loop
controller: in vivo feasibility study using a swine model. Anesthesiology.
2019;130(3):394–403. https:// doi. org/ 10. 1097/ aln. 00000 00000 002581.
62. Rinehart J, Ma M, Calderon MD, Cannesson M. Feasibility of automated
titration of vasopressor infusions using a novel closed-loop control-
ler. J Clin Monit Comput. 2018;32(1):5–11. https:// doi. org/ 10. 1007/
s10877- 017- 9981-6.
63. Rinehart J, Joosten A, Ma M, Calderon MD, Cannesson M. Closed-loop
vasopressor control: in-silico study of robustness against pharmacody-
namic variability. J Clin Monit Comput. 2019;33(5):795–802. https:// doi.
org/ 10. 1007/ s10877- 018- 0234-0.
64. Ren W, Chen J, Liu J, Fu Z, Yao Y, Chen X, et al. Feasibility of intelligent
drug control in the maintenance phase of general anesthesia based on
convolutional neural network. Heliyon. 2023;9(1):e12481. https:// doi. org/
10. 1016/j. heliy on. 2022. e12481.
65. Berggreen J, Johansson A, Jahr J, Möller S, Jansson T. Deep learning on
ultrasound images visualizes the femoral nerve with good precision.
Healthcare. 2023;11(2):184. https:// doi. org/ 10. 3390/ healt hcare 11020 184.
66. Bowness J, Macfarlane A, Noble J, Higham H, Burckett-St LD. Anaesthesia,
nerve blocks and artificial intelligence. Anaesthesia News. 2021. [cited
2023 Dec 26]. Available from: https:// anaes theti sts. org/ Home/ Resou
rces- publi catio ns/ Anaes thesia- News- magaz ine/ Anaes thesia- News- Digit
al- July- 2021/ Anaes thesia- nerve- blocks- and- artifi cial- intel ligen ce.
67. Bowness JS, Burckett-St Laurent D, Hernandez N, Keane PA, Lobo C,
Margetts S, et al. Assistive artificial intelligence for ultrasound image
interpretation in regional anaesthesia: an external validation study. Br J
Anaesth. 2023;130(2):217–25. https:// doi. org/ 10. 1016/j. bja. 2022. 06. 031.
68. In Chan JJ, Ma J, Leng Y, Tan KK, Tan CW, Sultana R, et al. Machine learning
approach to needle insertion site identification for spinal anesthesia in
obese patients. BMC Anesthesiol. 2021;21(1):246. https:// doi. org/ 10. 1186/
s12871- 021- 01466-8.
69. Pesteie M, Lessoway V, Abolmaesumi P, Rohling RN. Automatic Localiza-
tion of the Needle Target for Ultrasound-Guided Epidural Injections. IEEE
Trans Med Imaging. 2018;37(1):81–92. https:// doi. org/ 10. 1109/ tmi. 2017.
27391 10.
70. Xia M, Ma W, Zuo M, Deng X, Xue F, Battaglini D, et al. Expert consensus
on difficult airway assessment. Hepatobiliary Surg Nutr. 2023;12(4):545–
66. https:// doi. org/ 10. 21037/ hbsn- 23- 46.
71. Tavolara TE, Gurcan MN, Segal S, Niazi MKK. Identification of difficult to
intubate patients from frontal face images using an ensemble of deep
learning models. Comput Biol Med. 2021;136:104737. https:// doi. org/ 10.
1016/j. compb iomed. 2021. 104737.
72. Pei B, Jin C, Cao S, Ji N, Xia M, Jiang H. Geometric morphometrics and
machine learning from three-dimensional facial scans for difficult mask
ventilation prediction. Front Med. 2023;10:1203023. https:// doi. org/ 10.
3389/ fmed. 2023. 12030 23.
73. Matava C, Pankiv E, Raisbeck S, Caldeira M, Alam F. A convolutional neural
network for real time classification, identification, and labelling of vocal
cord and tracheal using laryngoscopy and bronchoscopy video. J Med
Syst. 2020;44(2):44. https:// doi. org/ 10. 1007/ s10916- 019- 1481-4.
74. Liu J, Xie Y, Shu X, Chen Y, Sun Y, Zhong K, et al. Value function assessment
to different RL algorithms for heparin treatment policy of patients with
sepsis in ICU. Artif Intell Med. 2024;147:102726. https:// doi. org/ 10. 1016/j.
artmed. 2023. 102726.
75. Anzabi Zadeh S, Street WN, Thomas BW. Optimizing warfarin dosing
using deep reinforcement learning. J Biomed Inform. 2023;137:104267.
https:// doi. org/ 10. 1016/j. jbi. 2022. 104267.
76. Bernstein DJ. Cryptographic Competitions. J Cryptol. 2023;37(1):7. https://
doi. org/ 10. 1007/ s00145- 023- 09467-1.
77. Li P, Lai J, Zhou D, Yang Y, Wu W, Fang J. Multi-authority anonymous
authentication with public accountability for incentive-based applica-
tions. Comput Netw. 2023;231:109828. https:// doi. org/ 10. 1016/j. comnet.
2023. 109828.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in pub-
lished maps and institutional affiliations.