Endrit Pajaziti’s research while affiliated with University College London and other places

What is this page?


This page lists works of an author who doesn't have a ResearchGate profile or hasn't added the works to their profile yet. It is automatically generated from public (personal) data to further our legitimate goal of comprehensive and accurate scientific recordkeeping. If you are this author and want this page removed, please let us know.

Publications (15)


Window ductus: Part aortopulmonary window and part patent ductus arteriosus
  • Article

October 2024

·

17 Reads

JTCVS Techniques

Pooja Shetty

·

Anusha Jegatheeswaran

·

Endrit Pajaziti

·

[...]

·




The model architecture of Image2Flow
The hybrid image and graph convolutional neural network architecture of Image2Flow. It takes as input a 3D cardiac MRI and a template volume mesh of a pulmonary artery. It outputs the patient-specific pulmonary artery mesh with associated pressure and flow at each node.
A schematic of the steps involved in point-point correspondent volume mesh generation
(A) raw patient-specific volume mesh created from manual segmentation, (B) initial template volume mesh creation, (C) point-point correspondent volume mesh generation by transforming the initial template, (D) final template volume mesh creation by averaging the point-point correspondent meshes of the training data. Red indicates non-corresponding meshes and blue represents corresponding meshes. The wireframe rendering denotes surface meshes, while the solid rendering denotes volume meshes.
Segmentation accuracy
The best, median and worst Image2Flow segmentations compared to ‘MeshDeformNet’ and a 3D UNet.
(A) MNAES values of the Image2Flow predictions compared to the ground truth on the test set (n = 15) for pressure and velocity. (B) Difference in MNAES values between CFDI2F and CFDDL-seg.
The best, median and worst blood pressure, and velocity predictions of Image2Flow by MNAEs
The size and positioning between the true and predicted meshes are to scale.

+6

Image2Flow: A proof-of-concept hybrid image and graph convolutional neural network for rapid patient-specific pulmonary artery segmentation and CFD flow field calculation from 3D cardiac MRI data
  • Article
  • Full-text available

June 2024

·

64 Reads

Computational fluid dynamics (CFD) can be used for non-invasive evaluation of hemodynamics. However, its routine use is limited by labor-intensive manual segmentation, CFD mesh creation, and time-consuming simulation. This study aims to train a deep learning model to both generate patient-specific volume-meshes of the pulmonary artery from 3D cardiac MRI data and directly estimate CFD flow fields. This proof-of-concept study used 135 3D cardiac MRIs from both a public and private dataset. The pulmonary arteries in the MRIs were manually segmented and converted into volume-meshes. CFD simulations were performed on ground truth meshes and interpolated onto point-point correspondent meshes to create the ground truth dataset. The dataset was split 110/10/15 for training, validation, and testing. Image2Flow, a hybrid image and graph convolutional neural network, was trained to transform a pulmonary artery template to patient-specific anatomy and CFD values, taking a specific inlet velocity as an additional input. Image2Flow was evaluated in terms of segmentation, and the accuracy of predicted CFD was assessed using node-wise comparisons. In addition, the ability of Image2Flow to respond to increasing inlet velocities was also evaluated. Image2Flow achieved excellent segmentation accuracy with a median Dice score of 0.91 (IQR: 0.86–0.92). The median node-wise normalized absolute error for pressure and velocity magnitude was 11.75% (IQR: 9.60–15.30%) and 9.90% (IQR: 8.47–11.90), respectively. Image2Flow also showed an expected response to increased inlet velocities with increasing pressure and velocity values. This proof-of-concept study has shown that it is possible to simultaneously perform patient-specific volume-mesh based segmentation and pressure and flow field estimation using Image2Flow. Image2Flow completes segmentation and CFD in ~330ms, which is ~5000 times faster than manual methods, making it more feasible in a clinical environment.

Download

Shape-driven deep neural networks for fast acquisition of aortic 3D pressure and velocity flow fields

April 2023

·

425 Reads

·

17 Citations

Computational fluid dynamics (CFD) can be used to simulate vascular haemodynamics and analyse potential treatment options. CFD has shown to be beneficial in improving patient outcomes. However, the implementation of CFD for routine clinical use is yet to be realised. Barriers for CFD include high computational resources, specialist experience needed for designing simulation set-ups, and long processing times. The aim of this study was to explore the use of machine learning (ML) to replicate conventional aortic CFD with automatic and fast regression models. Data used to train/test the model consisted of 3,000 CFD simulations performed on synthetically generated 3D aortic shapes. These subjects were generated from a statistical shape model (SSM) built on real patient-specific aortas (N = 67). Inference performed on 200 test shapes resulted in average errors of 6.01% ±3.12 SD and 3.99% ±0.93 SD for pressure and velocity, respectively. Our ML-based models performed CFD in ∼0.075 seconds (4,000x faster than the solver). This proof-of-concept study shows that results from conventional vascular CFD can be reproduced using ML at a much faster rate, in an automatic process, and with reasonable accuracy.


Fig. 8 Best and worst aorta and pulmonary artery predictions. Flow fields of pressure and velocity displayed
Automatic segmentation of the great arteries for computational hemodynamic assessment

November 2022

·

107 Reads

·

7 Citations

Journal of Cardiovascular Magnetic Resonance

Background Computational fluid dynamics (CFD) is increasingly used for the assessment of blood flow conditions in patients with congenital heart disease (CHD). This requires patient-specific anatomy, typically obtained from segmented 3D cardiovascular magnetic resonance (CMR) images. However, segmentation is time-consuming and requires expert input. This study aims to develop and validate a machine learning (ML) method for segmentation of the aorta and pulmonary arteries for CFD studies. Methods 90 CHD patients were retrospectively selected for this study. 3D CMR images were manually segmented to obtain ground-truth (GT) background, aorta and pulmonary artery labels. These were used to train and optimize a U-Net model, using a 70-10-10 train-validation-test split. Segmentation performance was primarily evaluated using Dice score. CFD simulations were set up from GT and ML segmentations using a semi-automatic meshing and simulation pipeline. Mean pressure and velocity fields across 99 planes along the vessel centrelines were extracted, and a mean average percentage error (MAPE) was calculated for each vessel pair (ML vs GT). A second observer (SO) segmented the test dataset for assessment of inter-observer variability. Friedman tests were used to compare ML vs GT, SO vs GT and ML vs SO metrics, and pressure/velocity field errors. Results The network’s Dice score (ML vs GT) was 0.945 (interquartile range: 0.929–0.955) for the aorta and 0.885 (0.851–0.899) for the pulmonary arteries. Differences with the inter-observer Dice score (SO vs GT) and ML vs SO Dice scores were not statistically significant for either aorta or pulmonary arteries ( p = 0.741, p = 0.061). The ML vs GT MAPEs for pressure and velocity in the aorta were 10.1% (8.5–15.7%) and 4.1% (3.1–6.9%), respectively, and for the pulmonary arteries 14.6% (11.5–23.2%) and 6.3% (4.3–7.9%), respectively. Inter-observer (SO vs GT) and ML vs SO pressure and velocity MAPEs were of a similar magnitude to ML vs GT ( p > 0.2). Conclusions ML can successfully segment the great vessels for CFD, with errors similar to inter-observer variability. This fast, automatic method reduces the time and effort needed for CFD analysis, making it more attractive for routine clinical use.


Automatic Segmentation of the Great Arteries for Computational Hemodynamic Assessment

September 2022

·

94 Reads

Background: Computational fluid dynamics (CFD) is increasingly used to assess blood flow conditions in patients with congenital heart disease (CHD). This requires patient-specific anatomy, usually obtained from segmented 3D cardiovascular magnetic resonance (CMR) images. However, segmentation is time-consuming and needs expert input. This study aims to develop and validate a machine learning (ML) method for segmentation of the aorta and pulmonary arteries (PAs) for CFD studies. Methods: 90 CHD patients were retrospectively selected for this study. 3D CMR images were manually segmented to obtain ground-truth (GT) background, aorta and PA labels. These were used to train and optimize a U-Net model. Segmentation performance was primarily evaluated using Dice score. CFD simulations were set up from GT and ML segmentations using a semi-automatic meshing and simulation pipeline. Pressure and velocity fields were computed, and a mean average percentage error (MAPE) was calculated for each vessel pair. A secondary observer (SO) segmented the test dataset to assess inter-observer variability. Friedman tests were used to compare segmentation metrics and flow field errors. Results: The model's Dice score (ML vs GT) was 0.945 for the aorta and 0.885 for the PAs. Differences with the inter-observer Dice score (SO vs GT) and ML vs SO Dice scores were not statistically significant for either aorta or PAs. The ML vs GT MAPEs for pressure and velocity in the aorta were 10.1% and 4.1% respectively, and for the PAs 14.6% and 6.3%, respectively. Inter-observer (SO vs GT) and ML vs SO pressure and velocity MAPEs were of a similar magnitude to ML vs GT. Conclusions: The proposed method can successfully segment the great vessels for CFD, with errors similar to inter-observer variability. This reduces the time and effort needed for CFD analysis, making it more attractive for routine clinical use.


Figure 5: The general sequential, fully-connected DNN set-up used to build both pressure and velocity predictors ('CFD vector'
Figure 7: Distribution of errors due to PCA. Left: subject with highest errors in pressure after reconstruction with 20 PCA modes (4.32%). Right: subject with highest errors in velocity after reconstruction with 55 PCA modes (4.93%).
Figure 9: Normalised pressure/velocity error computed on all test cases (n=200) then averaged node-wise. Errors are visualised by projecting values on the template (SSM average) mesh. Boxplots with median and interquartile ranges show the distribution of the same data (node-averaged errors).
Figure 12: Scatter plot showing the relationship between subject error (SE) and shape mode scores for all 200 test subjects.
Deep neural networks for fast acquisition of aortic 3D pressure and velocity flow fields

August 2022

·

248 Reads

Computational fluid dynamics (CFD) can be used to simulate vascular haemodynamics and analyse potential treatment options. CFD has shown to be beneficial in improving patient outcomes. However, the implementation of CFD for routine clinical use is yet to be realised. Barriers for CFD include high computational resources, specialist experience needed for designing simulation set-ups, and long processing times. The aim of this study was to explore the use of machine learning (ML) to replicate conventional aortic CFD with automatic and fast regression models. Data used to train/test the model comprised of 3,000 CFD simulations performed on synthetically generated 3D aortic shapes. These subjects were generated from a statistical shape model (SSM) built on real patient-specific aortas (N=67). Inference performed on 200 test shapes resulted in average errors of 6.01% +/-3.12 SD and 3.99% +/-0.93 SD for pressure and velocity, respectively. Our ML-based models performed CFD in ~0.075 seconds (4,000x faster than the solver). This study shows that results from conventional vascular CFD can be reproduced using ML at a much faster rate, in an automatic process, and with high accuracy.



Patient baseline characteristics
Agreement with chosen surgical strategy (Arterial Switch Operation: Yes vs No) according to different 3D tools: the tick () shows the cases when the surgeon correctly identified the need for ASO; the cross () shows the cases when the surgeon did not correctly identified the need for ASO.
Enhanced 3D visualization for planning biventricular repair of double outlet right ventricle: a pilot study on the advantages of virtual reality

October 2021

·

194 Reads

·

13 Citations

European Heart Journal - Digital Health

Aims We aim to determine any additional benefit of virtual reality (VR) experience if compared to conventional cross-sectional imaging and standard 3D modelling when deciding on surgical strategy in patients with complex double outlet right ventricle (DORV). Methods and results We retrospectively selected ten consecutive patients with DORV and complex interventricular communications, who underwent biventricular repair. An arterial switch operation (ASO) was part of the repair in three of those. CT or cardiac MRI images were used to reconstruct patient-specific 3D anatomies, which were then presented using different visualisation modalities: 3D pdf, 3D printed models, and VR models. Two experienced paediatric cardiac surgeons, blinded to repair performed, reviewed each case evaluating the suitability of repair following assessment of each visualization modalities. In addition, they had to identify those who had ASO as part of the procedure. Answers of the two surgeons were compared to the actual operations performed. There was no mortality during the follow-up (mean = 2.5 years). Two patients required reoperations. After review of CT/CMR images, the evaluators identified the surgical strategy in accordance with the actual surgical plan in 75% of the cases. When using 3D pdf this reached only 70%. Accordance improved to 85% after revision of 3D printed models and to 95% after VR. Use of 3D printed models and VR facilitated the identification of patients who required ASO. Conclusion VR can enhance understanding of suitability for biventricular repair in patients with complex DORV if compared to cross-sectional images and other 3D modelling techniques.


Citations (8)


... Accurate simulations demand high computational costs and time, unsuitable in clinical settings that require rapid decision-making [16,17]. To overcome this challenge, researchers have increasingly adopted Machine Learning (ML) alongside traditional CFD to push the boundaries of biomechanical research and applications [18][19][20][21][22][23][24][25][26][27][28][29][30][31]. These methods have proven effective in various haemodynamics studies, ranging from predicting blood flow quantities and haemodynamic indices [19][20][21][22][23][24][25][26] to enhancing flow data resolution and noise reduction [27][28][29][30][31]. ...

Reference:

ML-ROM Wall Shear Stress Prediction in Patient-Specific Vascular Pathologies under a Limited Clinical Training Data Regime
Shape-driven deep neural networks for fast acquisition of aortic 3D pressure and velocity flow fields

... Instead, we describe an alternative automatic 3D ML segmentation approach [8] that produces a high-resolution isotropic 3D segmentation of the thoracic aorta from a stack of 2D anisotropic localizers [9]. Our approach uses simulated 2D anisotropic localizers that are generated from 3D isotropic whole heart data. ...

Automatic segmentation of the great arteries for computational hemodynamic assessment

Journal of Cardiovascular Magnetic Resonance

... In the era of education informatization 2.0, the virtual practice teaching laboratory is a product of the deep integration of digital 3D technology and experimental teaching [16]. Qian, J. [17] selected 160 students from the same school for teaching experimental research and found that different virtual reality digital media art creation teaching methods based on artificial intelligence have the best effect. ...

Investigating the Feasibility of Virtual Reality for Teaching on Congenital Heart Disease
  • Citing Article
  • February 2022

European Journal of Vascular and Endovascular Surgery

... 3D prints can be used in discussing the diagnosis with both family/caretakers and other medical professionals and help to coordinate the surgical plan. Printed models can better identify suitable patients for biventricular repair and help in deciding upon the optimal surgical method [19]. Biventricular repair gives better post-operative outcomes and less reinterventions are necessary than with palliative univentricular repair [7]. ...

Enhanced 3D visualization for planning biventricular repair of double outlet right ventricle: a pilot study on the advantages of virtual reality

European Heart Journal - Digital Health

... Yaqoob introduced the combination of VR technology and short videos to explore the immersive experience effect of videos, providing reference for future research [9][10][11]. Scholars such as Gonzalez et al. have achieved automatic medical imaging segmentation through 3D modeling and the introduction of VR technology [12][13][14]. Kim et al. used 360°videos to investigate immersive VR content and explore its impact on students' learning outcomes [15]. In order to fully consider the real-time requirements of short videos, scholars such as Gionfrida et al. used a combination of 3D Convolutional Neural Network (3D CNN) and short-term memory units for real-time segmentation of gesture videos [16]. ...

Investigating the Feasibility of Virtual Reality (VR) for Teaching Cardiac Morphology

Electronics

... However, 72% of attendees found the methods of interaction (e.g. grabbing objects, using a cutting tool) "extremely intuitive'' (5/5) with 94% responding as very "very willing"(4/5) or "extremely willing"(5/5) to implement a VR setup at their home institutions [38]. ...

P369 Patient specific virtual reality for education in congenital heart disease

European Heart Journal Cardiovascular Imaging

... This echoes the finding from a recent meta-analysis, which found preoperative planning being the most relevant application of 3DPHM [24]. VR has also been reported in the current literature for its ability to provide an immersive, interactive, and free-form visualization experience, despite it not being tactile like 3DPHM [12,14,[25][26][27]. Unlike 3DPHM, being static and unable to show cardiac functional information, the VR project can be "programmed" to show dynamic cardiac models [28,29], to allow users to scale, rotate, crop the cardiac models, and change the viewing planes according to their needs [12]. ...

Taking surgery out of reality: A virtual journey into double outlet right ventricle

... Larger case series included XR for planning surgery in 5 cases of pulmonary atresia and major aortopulmonary collaterals (MAPCAs) [15], complex aortic valve reconstruction in 26 patients unsuitable for valve replacement [16], 17 pulmonary artery reconstructions and unifocalisations [17], and 6 complex redo and minimally invasive surgeries in adults [18]. In seven case reports XR was used for planning repair of DORV [19], truncus arteriosus [20], a complex ventricular septal defect (VSD) [21], right ventricular to pulmonary artery conduit [37]. Another group published a series of feasibility studies testing AR intra-procedural guidance for minimally-invasive cardiac surgery in phantom and animal models. ...

Taking Surgery Out of Reality

Circulation Cardiovascular Imaging