No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
A Novel Lightweight Deep Learning Framework with Knowledge Distillation for Efficient Diabetic Foot Ulcer Detection
... Amjad et al.[55] proposed a lightweight model called DFU-LWNet and trained it by InceptionV3 (teacher model) through knowledge distillation. The proposed model contains three convolutional layers with the max-pooling layers, and convolutional modules are borrowed from the Efficient-Net model. ...
Edge Machine Learning (EdgeML) and Tiny Machine Learning (TinyML) are fast-growing fields that bring machine learning to resource-constrained devices, allowing real-time data processing and decision-making at the network’s edge. However, the complexity of model conversion techniques, diverse inference mechanisms, and varied learning strategies make designing and deploying these models challenging. Additionally, deploying TinyML models on resource-constrained hardware with specific software frameworks has broadened EdgeML’s applications across various sectors. These factors underscore the necessity for a comprehensive literature review, as current reviews do not systematically encompass the most recent findings on these topics. Consequently, it provides a comprehensive overview of state-of-the-art techniques in model conversion, inference mechanisms, learning strategies within EdgeML, and deploying these models on resource-constrained edge devices using TinyML. It identifies 90 research articles published between 2018 and 2025, categorizing them into two main areas: (1) model conversion, inference, and learning strategies in EdgeML and (2) deploying TinyML models on resource-constrained hardware using specific software frameworks. In the first category, the synthesis of selected research articles compares and critically reviews various model conversion techniques, inference mechanisms, and learning strategies. In the second category, the synthesis identifies and elaborates on major development boards, software frameworks, sensors, and algorithms used in various applications across six major sectors. As a result, this article provides valuable insights for researchers, practitioners, and developers. It assists them in choosing suitable model conversion techniques, inference mechanisms, learning strategies, hardware development boards, software frameworks, sensors, and algorithms tailored to their specific needs and applications across various sectors.
Meeting the rising global demand for healthcare diagnostic tools is crucial, especially with a shortage of medical professionals. This issue has increased interest in utilizing deep learning (DL) and telemedicine technologies. DL, a branch of artificial intelligence, has progressed due to advancements in digital technology and data availability and has proven to be effective in solving previously challenging learning problems. Convolutional neural networks (CNNs) show potential in image detection and recognition, particularly in healthcare applications. However, due to their resource-intensiveness, they surpass the capabilities of general-purpose CPUs. Therefore, hardware accelerators such as application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), and graphics processing units (GPUs) have been developed. With their parallelism efficiency and energy-saving capabilities, FPGAs have gained popularity for DL networks. This research aims to automate the classification of normal and abnormal (specifically Diabetic Foot Ulcer—DFU) classes using various parallel hardware accelerators. The study introduces two CNN models, namely DFU_FNet and DFU_TFNet. DFU_FNet is a simple model that extracts features used to train classifiers like SVM and KNN. On the other hand, DFU_TFNet is a deeper model that employs transfer learning to test hardware efficiency on both shallow and deep models. DFU_TFNet has outperformed AlexNet, VGG16, and GoogleNet benchmarks with an accuracy 99.81%, precision 99.38% and F1-Score 99.25%. In addition, the study evaluated two high-performance computing platforms, GPUs and FPGAs, for real-time system requirements. The comparison of processing time and power consumption revealed that while GPUs outpace FPGAs in processing speed, FPGAs exhibit significantly lower power consumption than GPUs.
The diabetic foot ulcer (DFU) is a significant medical complication for diabetic patients, which often leads to lower limb amputation. The manual identification of ischaemia in DFU is laborious, time-consuming, and costly. This study aims to develop an efficient deep learning (DL) method for the early identification of ischaemia in DFU. Therefore, a novel Convolutional Neural Network (CNN) architecture (HCNNet) is proposed by integrating multiple hybridised blocks from inception, residual, and dense modules along with appropriately placed Squeeze-and-Excitation (SE) block and intermediate transition layers. The proposed HCNNet is trained several times using various optimizer and learning rate settings to optimise its performance. It achieves promising Area Under the ROC Curve (AUC) scores of 0.999 for ischaemia identification. The experimental results show that the proposed HCNNet outperforms existing State-Of-The-Art (SOTA) methods.
Recently, deep neural networks (DNNs) have been used successfully in many fields, particularly, in medical diagnosis. However, deep learning (DL) models are expensive in terms of memory and computing resources, which hinders their implementation in limited-resources devices or for delay-sensitive systems. Therefore, these deep models need to be accelerated and compressed to smaller sizes to be deployed on edge devices without noticeably affecting their performance. In this paper, recent accelerating and compression approaches of DNN are analyzed and compared regarding their performance, applications, benefits, and limitations with a more focus on the knowledge distillation approach as a successful emergent approach in this field. In addition, a framework is proposed to develop knowledge distilled DNN models that can be deployed on fog/edge devices for automatic disease diagnosis. To evaluate the proposed framework, two compressed medical diagnosis systems are proposed based on knowledge distillation deep neural models for both COVID-19 and Malaria. The experimental results show that these knowledge distilled models have been compressed by 18.4% and 15% of the original model and their responses accelerated by 6.14x and 5.86%, respectively, while there were no significant drop in their performance (dropped by 0.9% and 1.2%, respectively). Furthermore, the distilled models are compared with other pruned and quantized models. The obtained results revealed the superiority of the distilled models in terms of compression rates and response time.
Background:
This publication represents a scheduled update of the 2019 guidelines of the International Working Group of the Diabetic Foot (IWGDF) addressing the use of systems to classify foot ulcers in people with diabetes in routine clinical practice. The guidelines are based on a systematic review of the available literature that identified 28 classifications addressed in 149 articles and, subsequently, expert opinion using the GRADE methodology.
Methods:
First, we have developed a list of classification systems considered as being potentially adequate for use in a clinical setting, through the summary of judgements for diagnostic tests, focussing on the usability, accuracy and reliability of each system to predict ulcer-related complications as well as use of resources. Second, we have determined, following group debate and consensus, which of them should be used in specific clinical scenarios. Following this process, in a person with diabetes and a foot ulcer we recommend: (a) for communication among healthcare professionals: to use the SINBAD (Site, Ischaemia, Bacterial infection, Area and Depth) system (first option) or consider using WIfI (Wound, Ischaemia, foot Infection) system (alternative option, when the required equipment and level of expertise is available and it is considered feasible) and in each case the individual variables that compose the systems should be described rather than a total score; (b) for predicting the outcome of an ulcer in a specific individual: no existing system could be recommended; (c) for characterising a person with an infected ulcer: the use of the IDSA/IWGDF classification (first option) or consider using the WIfI system (alternative option, when the required equipment and level of expertise is available and it is considered as feasible); (d) for characterising a person with peripheral artery disease: consider using the WIfI system as a means to stratify healing likelihood and amputation risk; (e) for the audit of outcome(s) of populations: the use of the SINBAD score.
Conclusions:
For all recommendations made using GRADE, the certainty of evidence was judged, at best, as being low. Nevertheless, based on the rational application of current data this approach allowed the proposal of recommendations, which are likely to have clinical utility.
Background:
Classification and scoring systems can help both clinical management and audit the outcomes of routine care.
Aim:
This study aimed to assess published systems used to characterise ulcers in people with diabetes to determine which should be recommended to (a) aid communication between health professionals, (b) predict clinical outcome of individual ulcers, (c) characterise people with infection and/or peripheral arterial disease, and (d) audit to compare outcomes in different populations. This systematic review is part of the process of developing the 2023 guidelines to classify foot ulcers from the International Working Group on Diabetic Foot.
Methods:
We searched PubMed, Scopus and Web of Science for articles published up to December 2021 which evaluated the association, accuracy or reliability of systems used to classify ulcers in people with diabetes. Published classifications had to have been validated in populations of >80% of people with diabetes and a foot ulcer.
Results:
We found 28 systems addressed in 149 studies. Overall, the certainty of the evidence for each classification was low or very low, with 19 (68%) of the classifications being assessed by ≤ 3 studies. The most frequently validated system was the one from Meggitt-Wagner, but the articles validating this system focused mainly on the association between the different grades and amputation. Clinical outcomes were not standardized but included ulcer-free survival, ulcer healing, hospitalisation, limb amputation, mortality, and cost.
Conclusion:
Despite the limitations, this systematic review provided sufficient evidence to support recommendations on the use of six particular systems in specific clinical scenarios.
Background
In adults with diabetes, diabetic foot ulcer (DFU) and amputation are common and associated with significant morbidity and mortality.
Purpose
Identify tools predicting risk of DFU or amputation that are prognostically accurate and clinically feasible.
Methods
We searched for systematic reviews (SRs) of tools predicting DFU or amputation published in multiple databases from initiation to January, 2023. We assessed risk of bias (ROB) and provided a narrative review of reviews describing performance characteristics (calibration and discrimination) of prognostically accurate tools. For such tools, we additionally reviewed original studies to ascertain clinical applicability and usability (variables included, score calculation, and risk categorization).
Results
We identified 3 eligible SRs predicting DFU or amputation risk. Two recent SRs (2020 and 2021) were rated as moderate and low ROB respectively. Four risk prediction models – Boyko, Martins-Mendes (simplified), Martins-Mendes (original), and PODUS 2020 had good prognostic accuracy for predicting DFU or amputation over time horizons ranging from 1- to 5-years. PODUS 2020 predicts absolute average risk (e.g., 6% risk of DFU at 2 years) and consists of 3-binary variables with a simple, summative scoring (0–4) making it feasible for clinic use. The other 3 models categorize risk subjectively (e.g., high-risk for DFU at 3 years), include 2–7 variables, and require a calculation device. No data exist to inform rescreening intervals. Furthermore, the effectiveness of targeted interventions in decreasing incidence of DFU or amputation in response to prediction scores is unknown.
Conclusions
In this review of reviews, we identified 4 prognostically accurate models that predict DFU or amputation in persons with diabetes. The PODUS 2020 model, predicting absolute average DFU risk at 2 years, has the most favorable prognostic accuracy and is clinically feasible. Rescreening intervals and effectiveness of intervention based on prediction score are uncertain.
The most recurrent side effect of diabetes is diabetic foot ulcers and if unattended cause imputations. Diabetic feet affect 15 to 25% of diabetic people globally. Diabetes complications are due to less or no awareness of the consequences of diabetes among diabetic patients. Technology leveraging is an attempt to create distinct, affordable, and simple diabetic foot diagnostic strategies for patients and doctors. This work proposes early detection and prognosis of diabetic foot ulcers using the EfficientNet, a deep neural network model. EfficientNet is applied to an image set of 844-foot images, composed of healthy and diabetic ulcer feet. Better performance is obtained compared to earlier models using EfficientNet by carefully balancing network width, depth, and image resolution. The EfficientNet performed better compared to popular models like AlexNet, GoogleNet, VGG16, and VGG19. It gave maximum accuracy, f1-score, recall, and precision of 98.97%, 98%, 98%, and 99%, respectively.
italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Motivation:
Infection (bacteria in the wound) and ischemia (insufficient blood supply) in Diabetic Foot Ulcers (DFUs) increase the risk of limb amputation.
Goal:
To develop an image-based DFU infection and ischemia detection system that uses deep learning.
Methods:
The DFU dataset was augmented using geometric and color image operations, after which binary infection and ischemia classification was done using the EfficientNet deep learning model and a comprehensive set of baselines.
Results:
The EfficientNets model achieved 99% accuracy in ischemia classification and 98% in infection classification, outperforming ResNet and Inception (87% accuracy) and Ensemble CNN, the prior state of the art (Classification accuracy of 90% for ischemia 73% for infection). EfficientNets also classified test images in a fraction (10% to 50%) of the time taken by baseline models.
Conclusions
: This work demonstrates that EfficientNets is a viable deep learning model for infection and ischemia classification.
Diabetic foot ulcer (DFU) is one of the most serious and alarming diabetic complications, which often leads to high amputation rates in diabetic patients. Machine learning is a part of the field of artificial intelligence, which can automatically learn models from data and better inform clinical decision‐making. We aimed to develop an accurate and explainable prediction model to estimate the risk of in‐hospital amputation in patients with DFU. A total of 618 hospitalised patients with DFU were included in this study. The patients were divided into non‐amputation, minor amputation or major amputation group. Light Gradient Boosting Machine (LightGBM) and 5‐fold cross‐validation tools were used to construct a multi‐class classification model to predict the three outcomes of interest. In addition, we used the SHapley Additive exPlanations (SHAP) algorithm to interpret the predictions of the model. Our area under the receiver‐operating‐characteristic curve (AUC) demonstrated a 0.90, 0.85 and 0.86 predictive ability for non‐amputation, minor amputation and major amputation outcomes, respectively. Taken together, our data demonstrated that the developed explainable machine learning model provided accurate estimates of the amputation rate in patients with DFU during hospitalisation. Besides, the model could inform individualised analyses of the patients' risk factors.
Diabetes foot ulceration (DFU) and amputation are a cause of significant morbidity. The prevention of DFU may be achieved by the identification of patients at risk of DFU and the institution of preventative measures through education and offloading. Several studies have reported that thermogram images may help to detect an increase in plantar temperature prior to DFU. However, the distribution of plantar temperature may be heterogeneous, making it difficult to quantify and utilize to predict outcomes. We have compared a machine learning-based scoring technique with feature selection and optimization techniques and learning classifiers to several state-of-the-art Convolutional Neural Networks (CNNs) on foot thermogram images and propose a robust solution to identify the diabetic foot. A comparatively shallow CNN model, MobilenetV2 achieved an F1 score of ∼95% for a two-feet thermogram image-based classification and the AdaBoost Classifier used 10 features and achieved an F1 score of 97 %. A comparison of the inference time for the best-performing networks confirmed that the proposed algorithm can be deployed as a smartphone application to allow the user to monitor the progression of the DFU in a home setting.
Imaging techniques are used to capture anomalies of the human body. The captured images must be understood for diagnosis, prognosis and treatment planning of the anomalies. Medical image understanding is generally performed by skilled medical professionals. However, the scarce availability of human experts and the fatigue and rough estimate procedures involved with them limit the effectiveness of image understanding performed by skilled medical professionals. Convolutional neural networks (CNNs) are effective tools for image understanding. They have outperformed human experts in many image understanding tasks. This article aims to provide a comprehensive survey of applications of CNNs in medical image understanding. The underlying objective is to motivate medical image understanding researchers to extensively apply CNNs in their research and diagnosis. A brief introduction to CNNs has been presented. A discussion on CNN and its various award-winning frameworks have been presented. The major medical image understanding tasks, namely image classification, segmentation, localization and detection have been introduced. Applications of CNN in medical image understanding of the ailments of brain, breast, lung and other organs have been surveyed critically and comprehensively. A critical discussion on some of the challenges is also presented.
The image classification based on cloud computing suffers from difficult deployment as the network depth and data volume increase. Due to the depth of the model and the convolution process of each layer will produce a great amount of calculation, the GPU and storage performance of the device are extremely demanding, and the GPU and storage devices equipped on the embedded and mobile terminals cannot support large models. So it is necessary to compress the model so that the model can be deployed on these devices. Meanwhile, traditional compression based methods often miss many global features during the compression process, resulting in low classification accuracy. To solve the problem, this paper proposes a lightweight neural network model based on dilated convolution and depthwise separable convolution with twenty-nine layers for image classification. The proposed model employs the dilated convolution to expand the receptive field during the convolution process while maintaining the number of convolution parameters, which can extract more high-level global semantic features to improve the classification accuracy. Also, the depthwise separable convolution is applied to reduce the network parameters and computational complexity in convolution operations, which reduces the size of the network. The proposed model introduces three hyperparameters: width multiplier, image resolution, and dilated rate, to compress the network on the premise of ensuring accuracy. The experimental results show that compared with GoogleNet, the network proposed in this paper improves the classification accuracy by nearly 1%, and the number of parameters is reduced by 3.7 million.
Transfer learning aims at improving the performance of target learners on target domains by transferring the knowledge contained in different but related source domains. In this way, the dependence on a large number of target-domain data can be reduced for constructing target learners. Due to the wide application prospects, transfer learning has become a popular and promising area in machine learning. Although there are already some valuable and impressive surveys on transfer learning, these surveys introduce approaches in a relatively isolated way and lack the recent advances in transfer learning. Due to the rapid expansion of the transfer learning area, it is both necessary and challenging to comprehensively review the relevant studies. This survey attempts to connect and systematize the existing transfer learning research studies, as well as to summarize and interpret the mechanisms and the strategies of transfer learning in a comprehensive way, which may help readers have a better understanding of the current research status and ideas. Unlike previous surveys, this survey article reviews more than 40 representative transfer learning approaches, especially homogeneous transfer learning approaches, from the perspectives of data and model. The applications of transfer learning are also briefly introduced. In order to show the performance of different transfer learning models, over 20 representative transfer learning models are used for experiments. The models are performed on three different data sets, that is, Amazon Reviews, Reuters-21578, and Office-31, and the experimental results demonstrate the importance of selecting appropriate transfer learning models for different applications in practice.
Classification and scoring systems can help both clinical management and audit outcomes of routine care. The aim of this study was to assess published systems of diabetic foot ulcers (DFUs) to determine which should be recommended for a given clinical purpose. Published classifications had to have been validated in populations of > 75% people with diabetes and a foot ulcer. Each study was assessed for internal and external validity and reliability. Eight key factors associated with failure to heal were identified from large clinical series and each classification was scored on the number of these key factors included. Classifications were then arranged according to their proposed purpose into one or more of four groups: (a) aid communication between health professionals, (b) predict clinical outcome of individual ulcers, (c) aid clinical management decision making for an individual case, and (d) audit to compare outcome in different populations. Thirty‐seven classification systems were identified of which 18 were excluded for not being validated in a population of >75% DFUs. The included 19 classifications had different purposes and were derived from different populations. Only six were developed in multicentre studies, just 13 were externally validated, and very few had evaluated reliability.Classifications varied in the number (4 ‐ 30), and definition of individual items and the diagnostic tools required. Clinical outcomes were not standardized but included ulcer‐free survival, ulcer healing, hospitalization, limb amputation, mortality, and cost. Despite the limitations, there was sufficient evidence to make recommendations on the use of particular classifications for the indications listed above.
Recognition and analysis of Diabetic Foot Ulcers (DFU) using computerized methods is an emerging research area with the evolution of image-based machine learning algorithms. However, existing research focuses on the recognition, detection, and segmentation of the DFU as well as tissue classification of DFU such as granulation, slough. According to DFU medical classification systems, the presence of infection (bacteria in the wound) and ischaemia (inadequate blood supply) has important clinical implications for DFU assessment, which were used to predict the risk of amputation. In this work, we propose a new dataset and computer vision techniques to identify the presence of infection and ischaemia in DFU. This is the first time, a DFU dataset with ground truth labels of ischaemia and infection cases is introduced for research purposes. For the hand-crafted machine learning approach, we propose a new feature descriptor, namely Superpixel Color Descriptor. Then, we used the Ensemble Convolutional Neural Network (CNN) model for more effective recognition of ischaemia and infection. We propose to use a natural data-augmentation method, which identifies the region of interest on foot images and focuses on finding the salient features existing in this area. Finally, we evaluate the performance of our proposed techniques on binary classification, i.e. ischaemia versus non-ischaemia and infection versus non-infection. Overall, our proposed method performs better in the classification of ischaemia than infection. We found that our proposed Ensemble CNN deep learning algorithms performed better for both classification tasks than hand-crafted machine learning algorithms, with 90% accuracy in ischaemia classification and 73% in infection classification.
Abstract Deep convolutional neural networks have performed remarkably well on many Computer Vision tasks. However, these networks are heavily reliant on big data to avoid overfitting. Overfitting refers to the phenomenon when a network learns a function with very high variance such as to perfectly model the training data. Unfortunately, many application domains do not have access to big data, such as medical image analysis. This survey focuses on Data Augmentation, a data-space solution to the problem of limited data. Data Augmentation encompasses a suite of techniques that enhance the size and quality of training datasets such that better Deep Learning models can be built using them. The image augmentation algorithms discussed in this survey include geometric transformations, color space augmentations, kernel filters, mixing images, random erasing, feature space augmentation, adversarial training, generative adversarial networks, neural style transfer, and meta-learning. The application of augmentation methods based on GANs are heavily covered in this survey. In addition to augmentation techniques, this paper will briefly discuss other characteristics of Data Augmentation such as test-time augmentation, resolution impact, final dataset size, and curriculum learning. This survey will present existing methods for Data Augmentation, promising developments, and meta-level decisions for implementing Data Augmentation. Readers will understand how Data Augmentation can improve the performance of their models and expand limited datasets to take advantage of the capabilities of big data.
Diabetic Foot Ulcer (DFU) is the main complication of Diabetes, which, if not properly treated, may lead to amputation. One of the approaches of DFU treatment depends on the attentiveness of clinicians and patients. This treatment approach has drawbacks such as the high cost of the diagnosis as well as the length of treatment. Although this approach gives powerful results, the need for a remote, cost-effective, and convenient DFU diagnosis approach is urgent. In this paper, we introduce a new dataset of 754-ft images which contain healthy skin and skin with a diabetic ulcer from different patients. A novel Deep Convolutional Neural Network, DFU_QUTNet, is also proposed for the automatic classification of normal skin (healthy skin) class versus abnormal skin (DFU) class. Stacking more layers to a traditional Convolutional Neural Network to make it very deep does not lead to better performance, instead leading to worse performance due to the gradient. Therefore, our proposed DFU_QUTNet network is designed based on the idea of increasing the width of the network while keeping the depth compared to the state-of-the-art networks. Our network has been proven to be very beneficial for gradient propagation, as the error can be back-propagated through multiple paths. It also helps to combine different levels of features at each step of the network. Features extracted by DFU_QUTNet network are used to train Support Vector Machine (SVM) and K-Nearest Neighbors (KNN) classifiers. For the sake of comparison, we have fine-tuned then re-trained and tested three pre-trained deep learning networks (GoogleNet, VGG16, and AlexNet) for the same task. The proposed DFU_QUTNet network outperformed the state-of-the-art CNN networks by achieving the F1-score of 94.5%.
In a short period of time, many areas of science have made a sharp transition towards data-dependent methods. In some cases, this process has been enabled by simultaneous advances in data acquisition and the development of networked system technologies. This new situation is particularly clear in the life sciences, where data overabundance has sparked a flurry of new methodologies for data management and analysis. This can be seen as a perfect scenario for the use of machine learning and computational intelligence techniques to address problems in which more traditional data analysis approaches might struggle. But, this scenario also poses some serious challenges. One of them is model interpretability and explainability, especially for complex nonlinear models. In some areas such as medicine and health care, not addressing such challenge might seriously limit the chances of adoption, in real practice, of computer-based systems that rely on machine learning and computational intelligence methods for data analysis. In this paper, we reflect on recent investigations about the interpretability and explainability of machine learning methods and discuss their impact on medicine and health care. We pay specific attention to one of the ways in which interpretability and explainability in this context can be addressed, which is through data and model visualization. We argue that, beyond improving model interpretability as a goal in itself, we need to integrate the medical experts in the design of data analysis interpretation strategies. Otherwise, machine learning is unlikely to become a part of routine clinical and health care practice.
The lifetime risk for diabetic patients to develop a diabetic foot ulcer (DFU) is 25%. In these patients, the risk of amputation is increased and the outcome deteriorates.
More than 50% of non-traumatic lower-extremity amputations are related to DFU infections and 85% of all lower-extremity amputations in patients with diabetes are preceded by an ulcer; up to 70% of diabetic patients with a DFU-related amputation die within five years of their amputation.
Optimal management of patients with DFUs must include clinical awareness, adequate blood glucose control, periodic foot inspection, custom therapeutic footwear, off-loading in high-risk patients, local wound care, diagnosis and control of osteomyelitis and ischaemia.
Cite this article: EFORT Open Rev 2018;3:513-525. DOI: 10.1302/2058-5241.3.180010
Diabetic foot ulcers (DFUs) are a serious complication of diabetes that results in significant morbidity and mortality. Mortality rates associated with the development of a DFU are estimated to be 5% in the first 12 months, and 5-year morality rates have been estimated at 42%. The standard practices in DFU management include surgical debridement, dressings to facilitate a moist wound environment and exudate control, wound off-loading, vascular assessment, and infection and glycemic control. These practices are best coordinated by a multidisciplinary diabetic foot wound clinic. Even with this comprehensive approach, there is still room for improvement in DFU outcomes. Several adjuvant therapies have been studied to reduce DFU healing times and amputation rates. We reviewed the rationale and guidelines for current standard of care practices and reviewed the evidence for the efficacy of adjuvant agents. The adjuvant therapies reviewed include the following categories: nonsurgical debridement agents, dressings and topical agents, oxygen therapies, negative pressure wound therapy, acellular bioproducts, human growth factors, energy-based therapies, and systemic therapies. Many of these agents have been found to be beneficial in improving wound healing rates, although a large proportion of the data are small, randomized controlled trials with high risks of bias.
Globally, in 2016, one out of eleven adults suffered from Diabetes Mellitus. Diabetic Foot Ulcers (DFU) are a major complication of this disease, which if not managed properly can lead to amputation. Current clinical approaches to DFU treatment rely on patient and clinician vigilance, which has significant limitations such as the high cost involved in the diagnosis, treatment and lengthy care of the DFU. We collected an extensive dataset of foot images, which contain DFU from different patients. In this paper, we have proposed the use of traditional computer vision features for detecting foot ulcers among diabetic patients, which represent a cost-effective, remote and convenient healthcare solution. Furthermore, we used Convolutional Neural Networks (CNNs) for the first time in DFU classification. We have proposed a novel convolutional neural network architecture, DFUNet, with better feature extraction to identify the feature differences between healthy skin and the DFU. Using 10-fold cross-validation, DFUNet achieved an AUC score of 0.962. This outperformed both the machine learning and deep learning classifiers we have tested. Here we present the development of a novel and highly sensitive DFUNet for objectively detecting the presence of DFUs. This novel approach has the potential to deliver a paradigm shift in diabetic foot care.
Recent work has shown that convolutional networks can be substantially deeper, more accurate and efficient to train if they contain shorter connections between layers close to the input and those close to the output. In this paper we embrace this observation and introduce the Dense Convolutional Network (DenseNet), where each layer is directly connected to every other layer in a feed-forward fashion. Whereas traditional convolutional networks with L layers have L connections, one between each layer and its subsequent layer (treating the input as layer 0), our network has L(L+1)/2 direct connections. For each layer, the feature maps of all preceding layers are treated as separate inputs whereas its own feature maps are passed on as inputs to all subsequent layers. Our proposed connectivity pattern has several compelling advantages: it alleviates the vanishing gradient problem and strengthens feature propagation; despite the increase in connections, it encourages feature reuse and leads to a substantial reduction of parameters; its models tend to generalize surprisingly well. We evaluate our proposed architecture on five highly competitive object recognition benchmark tasks. The DenseNet obtains significant improvements over the state-of-the-art on all five of them (e.g., yielding 3.74% test error on CIFAR-10, 19.25% on CIFAR-100 and 1.59% on SVHN).
Deep learning algorithms, in particular convolutional networks, have rapidly become a methodology of choice for analyzing medical images. This paper reviews the major deep learning concepts pertinent to medical image analysis and summarizes over 300 contributions to the field, most of which appeared in the last year. We survey the use of deep learning for image classification, object detection, segmentation, registration, and other tasks and provide concise overviews of studies per application area. Open challenges and directions for future research are discussed.
Recent work has shown that convolutional networks can be substantially deeper, more accurate, and efficient to train if they contain shorter connections between layers close to the input and those close to the output. In this paper, we embrace this observation and introduce the Dense Convolutional Network (DenseNet), which connects each layer to every other layer in a feed-forward fashion. Whereas traditional convolutional networks with L layers have L connections - one between each layer and its subsequent layer - our network has L(L+1)/2 direct connections. For each layer, the feature-maps of all preceding layers are used as inputs, and its own feature-maps are used as inputs into all subsequent layers. DenseNets have several compelling advantages: they alleviate the vanishing-gradient problem, strengthen feature propagation, encourage feature reuse, and substantially reduce the number of parameters. We evaluate our proposed architecture on four highly competitive object recognition benchmark tasks (CIFAR-10, CIFAR-100, SVHN, and ImageNet). DenseNets obtain significant improvements over the state-of-the-art on most of them, whilst requiring less memory and computation to achieve high performance. Code and models are available at https://github.com/liuzhuang13/DenseNet.
This is a post hoc analysis of quality of life in diabetic neuropathy patients in a cross-sectional survey performed in 2012 in Romania, using the Norfolk QOL-DN in which 21,756 patients with self-reported diabetes were enrolled. This current analysis aims to expand research on the diabetic foot and to provide an update on the number of foot ulcers found in Romania. Of the 21,174 patients included in this analysis, 14.85% reported a history of foot ulcers and 3.60% reported an amputation. The percentage of neuropathy patients with foot ulcers increased with age; the lowest percentage was observed in the 20–29-year age group (6.62%) and the highest in the 80–89-year age group (17.68%). The highest number of amputations was reported in the 70–79-year age group (largest group). Compared to patients without foot ulcers, those with foot ulcers had significantly higher scores for total DN and all its subdomains translating to worse QOL (
p
<
0.001
). This analysis showed a high rate of foot ulcers and amputations in Romanian diabetic patients. It underscores the need for implementation of effective screening and educational programs.
This paper compares well-established Convolutional Neural Networks (CNNs) to recently introduced Vision Transformers for the task of Diabetic Foot Ulcer Classification, in the context of the DFUC 2021 Grand-Challenge, in which this work attained the first position. Comprehensive experiments demonstrate that modern CNNs are still capable of outperforming Transformers in a low-data regime, likely owing to their ability for better exploiting spatial correlations. In addition, we empirically demonstrate that the recent Sharpness-Aware Minimization (SAM) optimization algorithm improves considerably the generalization capability of both kinds of models. Our results demonstrate that for this task, the combination of CNNs and the SAM optimization process results in superior performance than any other of the considered approaches.KeywordsDiabetic Foot Ulcer ClassificationVision TransformersConvolutional Neural NetworksSharpness-Aware Optimization
Background and objectives
The traditional method of detecting COVID-19 disease mainly rely on the interpretation of computer tomography (CT) or X-ray images (X-ray) by doctors or professional researchers to identify whether it is COVID-19 disease, which is easy to cause identification mistakes. In this study, the technology of convolutional neural network is expected to be able to efficiently and accurately identify the COVID-19 disease.
Methods
This study uses and fine-tunes seven convolutional neural networks including InceptionV3, ResNet50V2, Xception, DenseNet121, MobileNetV2, EfficientNet-B0, and EfficientNetV2 on COVID-19 detection. In addition, we proposes a lightweight convolutional neural network, LightEfficientNetV2, on small number of chest X-ray and CT images. Five-fold cross-validation was used to evaluate the performance of each model. To confirm the performance of the proposed model, LightEfficientNetV2 was carried out on three different datasets (NIH Chest X-rays, SARS-CoV-2 and COVID-CT).
Results
On chest X-ray image dataset, the highest accuracy 96.50% was from InceptionV3 before fine-tuning; and the highest accuracy 97.73% was from EfficientNetV2 after fine-tuning. The accuracy of the LightEfficientNetV2 model proposed in this study is 98.33% on chest X-ray image. On CT images, the best transfer learning model before fine-tuning is MobileNetV2, with an accuracy of 94.46%; the best transfer learning model after fine-tuning is Xception, with an accuracy of 96.78%. The accuracy of the LightEfficientNetV2 model proposed in this study is 97.48% on CT image.
Conclusions
Compared with the SOTA, LightEfficientNetV2 proposed in this study demonstrates promising performance on chest X-ray images, CT images and three different datasets.
2018 Curran Associates Inc.All rights reserved. Batch Normalization (BatchNorm) is a widely adopted technique that enables faster and more stable training of deep neural networks (DNNs). Despite its pervasiveness, the exact reasons for BatchNorm's effectiveness are still poorly understood. The popular belief is that this effectiveness stems from controlling the change of the layers' input distributions during training to reduce the so-called “internal covariate shift”. In this work, we demonstrate that such distributional stability of layer inputs has little to do with the success of BatchNorm. Instead, we uncover a more fundamental impact of BatchNorm on the training process: it makes the optimization landscape significantly smoother. This smoothness induces a more predictive and stable behavior of the gradients, allowing for faster training.
There has been a substantial amount of research involving computer methods and technology for the detection and recognition of diabetic foot ulcers (DFUs), but there is a lack of systematic comparisons of state-of-the-art deep learning object detection frameworks applied to this problem. DFUC2020 provided participants with a comprehensive dataset consisting of 2,000 images for training and 2,000 images for testing. This paper summarizes the results of DFUC2020 by comparing the deep learning-based algorithms proposed by the winning teams: Faster R–CNN, three variants of Faster R–CNN and an ensemble method; YOLOv3; YOLOv5; EfficientDet; and a new Cascade Attention Network. For each deep learning method, we provide a detailed description of model architecture, parameter settings for training and additional stages including pre-processing, data augmentation and post-processing. We provide a comprehensive evaluation for each method. All the methods required a data augmentation stage to increase the number of images available for training and a post-processing stage to remove false positives. The best performance was obtained from Deformable Convolution, a variant of Faster R–CNN, with a mean average precision (mAP) of 0.6940 and an F1-Score of 0.7434. Finally, we demonstrate that the ensemble method based on different deep learning methods can enhance the F1-Score but not the mAP.
Acute and chronic wounds are a challenge to healthcare systems around the world and affect many people’s lives annually. Wound classification is a key step in wound diagnosis that would help clinicians to identify an optimal treatment procedure. Hence, having a high-performance classifier assists wound specialists to classify wound types with less financial and time costs. Different wound classification methods based on machine learning and deep learning have been proposed in the literature. In this study, we have developed an ensemble Deep Convolutional Neural Network-based classifier to categorize wound images into multiple classes including surgical, diabetic, and venous ulcers. The output classification scores of two classifiers (namely, patch-wise and image-wise) are fed into a Multilayer Perceptron to provide a superior classification performance. A 5-fold cross-validation approach is used to evaluate the proposed method. We obtained maximum and average classification accuracy values of 96.4% and 94.28% for binary and 91.9% and 87.7% for 3-class classification problems. The proposed classifier was compared with some common deep classifiers and showed significantly higher accuracy metrics. We also tested the proposed method on the Medetec wound image dataset, and the accuracy values of 91.2% and 82.9% were obtained for binary and 3-class classifications. The results show that our proposed method can be used effectively as a decision support system in classification of wound images or other related clinical applications.
Background:
Diabetes mellitus, a metabolic disorder characterised by hyperglycaemia and associated with a heavy burden of microvascular and macrovascular complications, frequently remains undiagnosed. Screening of apparently healthy individuals may lead to early detection and treatment of type 2 diabetes mellitus and may prevent or delay the development of related complications.
Objectives:
To assess the effects of screening for type 2 diabetes mellitus.
Search methods:
We searched CENTRAL, MEDLINE, LILACS, the WHO ICTRP, and ClinicalTrials.gov from inception. The date of the last search was May 2019 for all databases. We applied no language restrictions.
Selection criteria:
We included randomised controlled trials involving adults and children without known diabetes mellitus, conducted over at least three months, that assessed the effect of diabetes screening (mass, targeted, or opportunistic) compared to no diabetes screening.
Data collection and analysis:
Two review authors independently screened titles and abstracts for potential relevance and reviewed the full-texts of potentially relevant studies, extracted data, and carried out 'Risk of bias' assessment using the Cochrane 'Risk of bias' tool. We assessed the overall certainty of the evidence using the GRADE approach.
Main results:
We screened 4651 titles and abstracts identified by the search and assessed 92 full-texts/records for inclusion. We included one cluster-randomised trial, the ADDITION-Cambridge study, which involved 20,184 participants from 33 general practices in Eastern England and assessed the effects of inviting versus not inviting high-risk individuals to screening for diabetes. The diabetes risk score was used to identify high-risk individuals; it comprised variables relating to age, sex, body mass index, and the use of prescribed steroid and anti-hypertensive medication. Twenty-seven practices were randomised to the screening group (11,737 participants actually attending screening) and 5 practices to the no-screening group (4137 participants). In both groups, 36% of participants were women; the average age of participants was 58.2 years in the screening group and 57.9 years in the no-screening group. Almost half of participants in both groups were on antihypertensive medication. The findings from the first phase of this study indicate that screening compared to no screening for type 2 diabetes did not show a clear difference in all-cause mortality (hazard ratio (HR) 1.06, 95% confidence interval (CI) 0.90 to 1.25, low-certainty evidence). Screening compared to no screening for type 2 diabetes mellitus showed an HR of 1.26, 95% CI 0.75 to 2.12 (low-certainty evidence) for diabetes-related mortality (based on whether diabetes was reported as a cause of death on the death certificate). Diabetes-related morbidity and health-related quality of life were only reported in a subsample and did not show a substantial difference between the screening intervention and control. The included study did not report on adverse events, incidence of type 2 diabetes, glycosylated haemoglobin A1c (HbA1c), and socioeconomic effects.
Authors' conclusions:
We are uncertain about the effects of screening for type 2 diabetes on all-cause mortality and diabetes-related mortality. Evidence was available from one study only. We are therefore unable to draw any firm conclusions relating to the health outcomes of early type 2 diabetes mellitus screening. Furthermore, the included study did not assess all of the outcomes prespecified in the review (diabetes-related morbidity, incidence of type 2 diabetes, health-related quality of life, adverse events, socioeconomic effects).
Developing state-of-the-art image classification models often requires significant architecture engineering and tuning. In this paper, we attempt to reduce the amount of architecture engineering by using Neural Architecture Search to learn an architectural building block on a small dataset that can be transferred to a large dataset. This approach is similar to learning the structure of a recurrent cell within a recurrent network. In our experiments, we search for the best convolutional cell on the CIFAR-10 dataset and then apply this learned cell to the ImageNet dataset by stacking together more of this cell. Although the cell is not learned directly on ImageNet, an architecture constructed from the best learned cell achieves state-of-the-art accuracy of 82.3% top-1 and 96.0% top-5 on ImageNet, which is 0.8% better in top-1 accuracy than the best human-invented architectures while having 9 billion fewer FLOPS. This cell can also be scaled down two orders of magnitude: a smaller network constructed from the best cell also achieves 74% top-1 accuracy, which is 3.1% better than the equivalently-sized, state-of-the-art models for mobile platforms.
Lower-extremity complications of diabetes such as foot ulcers constitute a substantial burden for people with diabetes. Once healed, foot ulcers frequently recur. This fact, coupled with demographic trends, requires a collective refocusing on prevention and a reallocation of resources from simply healing active ulcers to maximizing ulcer-free days for all patients with a history of diabetic foot ulceration. Aggressive therapy during active disease combined with a focus on improving care during remission can lead to more ulcer-free days, fewer inpatient and outpatient visits, and an improved quality of life.
Convolutional networks are at the core of most stateof-the-art computer vision solutions for a wide variety of tasks. Since 2014 very deep convolutional networks started to become mainstream, yielding substantial gains in various benchmarks. Although increased model size and computational cost tend to translate to immediate quality gains for most tasks (as long as enough labeled data is provided for training), computational efficiency and low parameter count are still enabling factors for various use cases such as mobile vision and big-data scenarios. Here we are exploring ways to scale up networks in ways that aim at utilizing the added computation as efficiently as possible by suitably factorized convolutions and aggressive regularization. We benchmark our methods on the ILSVRC 2012 classification challenge validation set demonstrate substantial gains over the state of the art: 21.2% top-1 and 5.6% top-5 error for single frame evaluation using a network with a computational cost of 5 billion multiply-adds per inference and with using less than 25 million parameters. With an ensemble of 4 models and multi-crop evaluation, we report 3.5% top-5 error and 17.3% top-1 error.
Deep neural nets with a large number of parameters are very powerful machine learning systems. However, overfitting is a serious problem in such networks. Large networks are also slow to use, making it difficult to deal with overfitting by combining the predictions of many different large neural nets at test time. Dropout is a technique for addressing this problem. The key idea is to randomly drop units (along with their connections) from the neural network during training. This prevents units from co-adapting too much. During training, dropout samples from an exponential number of different "thinned" networks. At test time, it is easy to approximate the effect of averaging the predictions of all these thinned networks by simply using a single unthinned network that has smaller weights. This significantly reduces overfitting and gives major improvements over other regularization methods. We show that dropout improves the performance of neural networks on supervised learning tasks in vision, speech recognition, document classification and computational biology, obtaining state-of-the-art results on many benchmark data sets. © 2014 Nitish Srivastava, Geoffrey Hinton, Alex Krizhevsky, Ilya Sutskever and Ruslan Salakhutdinov.
Deeper neural networks are more difficult to train. We present a residual
learning framework to ease the training of networks that are substantially
deeper than those used previously. We explicitly reformulate the layers as
learning residual functions with reference to the layer inputs, instead of
learning unreferenced functions. We provide comprehensive empirical evidence
showing that these residual networks are easier to optimize, and can gain
accuracy from considerably increased depth. On the ImageNet dataset we evaluate
residual nets with a depth of up to 152 layers---8x deeper than VGG nets but
still having lower complexity. An ensemble of these residual nets achieves
3.57% error on the ImageNet test set. This result won the 1st place on the
ILSVRC 2015 classification task. We also present analysis on CIFAR-10 with 100
and 1000 layers.
The depth of representations is of central importance for many visual
recognition tasks. Solely due to our extremely deep representations, we obtain
a 28% relative improvement on the COCO object detection dataset. Deep residual
nets are foundations of our submissions to ILSVRC & COCO 2015 competitions,
where we also won the 1st places on the tasks of ImageNet detection, ImageNet
localization, COCO detection, and COCO segmentation.
Foot ulcers and their complications are an important cause of morbidity and mortality in diabetes. The present study aims to examine the long-term outcome in terms of amputations and mortality in patients with new-onset diabetic foot ulcers in subgroups stratified by etiology.
Patients presenting with new ulcers (duration <1 month) to a dedicated diabetic foot clinic between 1994 and 1998 were studied. Outcomes were determined until March 2000 (or death) from podiatry, hospital, and district registers. Baseline clinical examination was done to classify ulcers as neuropathic, ischemic, or neuroischemic. Five-year amputation and mortality rates were derived from Kaplan-Meier survival analysis curves.
Of the 185 patients studied, 41% had peripheral vascular disease (PVD) and 61% had neuropathy; 45%, 16%, and 24% of patients had neuropathic, ischemic, and neuroischemic ulcers, respectively. The mean follow-up period was 34 months (range 1-65) including survivors and patients who died during the study period. Five-year amputation rates were higher for ischemic (29%) and neuroischemic (25%) than neuropathic (11%) ulcers. Five-year mortality was 45%, 18%, and 55% for neuropathic, neuroischemic, and ischemic ulcers, respectively. Mortality was higher in ischemic ulcers than neuropathic ulcers. On multivariate regression analysis, only increasing age predicted shorter survival time.
All types of diabetic foot ulcers are associated with high morbidity and mortality. The increased mortality appears to be independent of factors increasing ulcer risk-that is, neuropathy and PVD-in patients with established foot ulcers.
Efficientnet: Rethinking model scaling for convolutional neural networks
- Tan
Hardnet-dfus: Enhancing backbone and decoder of hardnet-mseg for diabetic foot ulcer image segmentation
- Liao