Directory

Elif Keles | Northwestern University Feinberg School of Medicine - Academia.edu

Skip to main content

Elif Keles

Northwestern University Feinberg School of Medicine, Radiology, Post-Doc

Followers

16

Following

12

Public Views

Colleges of Applied Sciences

Vrije Universiteit Brussel

May Dongmei Wang

Joon Huang Chuah

University of Cambridge

Riries Rulaningtyas

Ishan Nedumkunnel .

InterestsView All (9)

Uploads

Papers by Elif Keles

Information Bottleneck Attribution for Visual Explanations of Diagnosis and Prognosis

arXiv (Cornell University), Apr 6, 2021

Visual explanation methods have an important role in the prognosis of the patients where the anno... more Visual explanation methods have an important role in the prognosis of the patients where the annotated data is limited or unavailable. There have been several attempts to use gradient-based attribution methods to localize pathology from medical scans without using segmentation labels. This research direction has been impeded by the lack of robustness and reliability. These methods are highly sensitive to the network parameters. In this study, we introduce a robust visual explanation method to address this problem for medical applications. We provide an innovative visual explanation algorithm for general purpose and as an example application we demonstrate its effectiveness for quantifying lesions in the lungs caused by the Covid-19 with high accuracy and robustness without using dense segmentation labels. This approach overcomes the drawbacks of commonly used Grad-CAM and its extended versions. The premise behind our proposed strategy is that the information flow is minimized while ensuring the classifier prediction stays similar. Our findings indicate that the bottleneck condition provides a more stable severity estimation than the similar attribution methods. The source code will be publicly available upon publication.

Selecting the best optimizers for deep learning–based medical image segmentation

Frontiers in Radiology

PurposeThe goal of this work is to explore the best optimizers for deep learning in the context o... more PurposeThe goal of this work is to explore the best optimizers for deep learning in the context of medical image segmentation and to provide guidance on how to design segmentation networks with effective optimization strategies.ApproachMost successful deep learning networks are trained using two types of stochastic gradient descent (SGD) algorithms: adaptive learning and accelerated schemes. Adaptive learning helps with fast convergence by starting with a larger learning rate (LR) and gradually decreasing it. Momentum optimizers are particularly effective at quickly optimizing neural networks within the accelerated schemes category. By revealing the potential interplay between these two types of algorithms [LR and momentum optimizers or momentum rate (MR) in short], in this article, we explore the two variants of SGD algorithms in a single setting. We suggest using cyclic learning as the base optimizer and integrating optimal values of learning rate and momentum rate. The new optimi...

A review of deep learning and radiomics approaches for pancreatic cancer diagnosis from medical imaging

Current Opinion in Gastroenterology

Purpose of review Early and accurate diagnosis of pancreatic cancer is crucial for improving pati... more Purpose of review Early and accurate diagnosis of pancreatic cancer is crucial for improving patient outcomes, and artificial intelligence (AI) algorithms have the potential to play a vital role in computer-aided diagnosis of pancreatic cancer. In this review, we aim to provide the latest and relevant advances in AI, specifically deep learning (DL) and radiomics approaches, for pancreatic cancer diagnosis using cross-sectional imaging examinations such as computed tomography (CT) and magnetic resonance imaging (MRI). Recent findings This review highlights the recent developments in DL techniques applied to medical imaging, including convolutional neural networks (CNNs), transformer-based models, and novel deep learning architectures that focus on multitype pancreatic lesions, multiorgan and multitumor segmentation, as well as incorporating auxiliary information. We also discuss advancements in radiomics, such as improved imaging feature extraction, optimized machine learning classif...

Tu1992 AI EMPOWERED AUTOMATIC VOLUME DELINEATION OF LIVER FROM CT SCANS FOR DIAGNOSTIC WORKFLOW

Gastroenterology

Tu1986 A FULLY AUTOMATIC AI SYSTEM FOR PANCREAS SEGMENTATION FROM MULTICENTER MRI SCANS

Gastroenterology

Ensuring Trustworthy Medical Artificial Intelligencethrough Ethical and Philosophical Principles

arXiv (Cornell University), Apr 23, 2023

Artificial Intelligence (AI) methods hold immense potential to revolutionize numerous medical car... more Artificial Intelligence (AI) methods hold immense potential to revolutionize numerous medical care by enhancing the experience of medical experts and patients. AI-based computer-assisted diagnosis and treatment tools can democratize healthcare by matching the clinical level or surpassing clinical experts. As a result, advanced healthcare services can be affordable to all populations, irrespective of demographics, race, or socioeconomic background. The democratization of such AI tools can reduce the cost of care, optimize resource allocation, and improve the quality of care. In contrast to humans, AI can uncover complex relations in the data from a large set of inputs and even lead to new evidence-based knowledge in medicine. However, integrating AI into healthcare raises several ethical and philosophical concerns, such as bias, transparency, autonomy, responsibility, and accountability. Here, we emphasize recent advances in AI-assisted medical image analysis, existing standards, and the significance of comprehending ethical issues and best practices for clinical settings. We cover the technical and ethical challenges and implications of deploying AI in hospitals and public organizations. We also discuss key measures and techniques to address ethical challenges, data scarcity, racial bias, lack of transparency, and algorithmic bias and provide recommendations and future directions.

Building Blocks of AI

AI in Clinical Medicine, Mar 3, 2023

Asphyxia Associated Metabolite Biomarker Investigation (AAMBI). Ergebnisse im Neugeborenenalter und mit 22-42 Monaten

30. Kongress der Deutschen Gesellschaft für Perinatale Medizin – „Wandel als Herausforderung“, 2021

Mortality from gastrointestinal congenital anomalies at 264 hospitals in 74 low-income, middle-income, and high-income countries: a multicentre, international, prospective cohort study

The Lancet, 2021

Background Congenital anomalies are the fifth leading cause of mortality in children younger than... more Background Congenital anomalies are the fifth leading cause of mortality in children younger than 5 years globally. Many gastrointestinal congenital anomalies are fatal without timely access to neonatal surgical care, but few studies have been done on these conditions in low-income and middle-income countries (LMICs). We compared outcomes of the seven most common gastrointestinal congenital anomalies in low-income, middle-income, and high-income countries globally, and identified factors associated with mortality. Methods We did a multicentre, international prospective cohort study of patients younger than 16 years, presenting to hospital for the first time with oesophageal atresia, congenital diaphragmatic hernia, intestinal atresia, gastroschisis, exomphalos, anorectal malformation, and Hirschsprung's disease. Recruitment was of consecutive patients for a minimum of 1 month between October, 2018, and April, 2019. We collected data on patient demographics, clinical status, interventions, and outcomes using the REDCap platform. Patients were followed up for 30 days after primary intervention, or 30 days after admission if they did not receive an intervention. The primary outcome was all-cause, in-hospital mortality for all conditions combined and each condition individually, stratified by country income status. We did a complete case analysis. Findings We included 3849 patients with 3975 study conditions (560 with oesophageal atresia, 448 with congenital diaphragmatic hernia, 681 with intestinal atresia, 453 with gastroschisis, 325 with exomphalos, 991 with anorectal malformation, and 517 with Hirschsprung's disease) from 264 hospitals (89 in high-income countries, 166 in middleincome countries, and nine in low-income countries) in 74 countries. Of the 3849 patients, 2231 (58•0%) were male. Median gestational age at birth was 38 weeks (IQR 36-39) and median bodyweight at presentation was 2•8 kg (2•3-3•3). Mortality among all patients was 37 (39•8%) of 93 in low-income countries, 583 (20•4%) of 2860 in middle-income countries, and 50 (5•6%) of 896 in high-income countries (p<0•0001 between all country income groups). Gastroschisis had the greatest difference in mortality between country income strata (nine [90•0%] of ten in lowincome countries, 97 [31•9%] of 304 in middle-income countries, and two [1•4%] of 139 in high-income countries; p≤0•0001 between all country income groups). Factors significantly associated with higher mortality for all patients combined included country income status (low-income vs high-income countries, risk ratio 2•78 [95% CI 1•88-4•11], p<0•0001; middle-income vs high-income countries, 2•11 [1•59-2•79], p<0•0001), sepsis at presentation (1•20 [1•04-1•40], p=0•016), higher American Society of Anesthesiologists (ASA) score at primary intervention (ASA 4-5 vs ASA 1-2, 1•82 [1•40-2•35], p<0•0001; ASA 3 vs ASA 1-2, 1•58, [1•30-1•92], p<0•0001]), surgical safety checklist not used (1•39 [1•02-1•90], p=0•035), and ventilation or parenteral nutrition unavailable when needed (ventilation 1•96, [1•41-2•71], p=0•0001; parenteral nutrition 1•35, [1•05-1•74], p=0•018). Administration of parenteral nutrition (0•61, [0•47-0•79], p=0•0002) and use of a peripherally inserted central catheter (0•65 [0•50-0•86], p=0•0024) or percutaneous central line (0•69 [0•48-1•00], p=0•049) were associated with lower mortality. Interpretation Unacceptable differences in mortality exist for gastrointestinal congenital anomalies between lowincome, middle-income, and high-income countries. Improving access to quality neonatal surgical care in LMICs will be vital to achieve Sustainable Development Goal 3.2 of ending preventable deaths in neonates and children younger than 5 years by 2030. Funding Wellcome Trust.

Die AAMBI-Studie (Asphyxia Associated Metabolite Biomarker Investigation): Ergebnisse im Neugeborenenalter und mit 22–42 Monaten

Zeitschrift für Geburtshilfe und Neonatologie

Deep Learning Algorithms for Pancreas Segmentation from Radiology Scans

Advances in Clinical Radiology

Selecting the Best Optimizers for Deep Learning based Medical Image Segmentation

arXiv (Cornell University), Feb 4, 2023

The goal of this work is to identify the best optimizers for deep learning in the context of card... more The goal of this work is to identify the best optimizers for deep learning in the context of cardiac image segmentation and to provide guidance on how to design segmentation networks with effective optimization strategies. Approach: Most successful deep learning networks are trained using two types of stochastic gradient descent (SGD) algorithms: adaptive learning and accelerated schemes. Adaptive learning helps with fast convergence by starting with a larger learning rate (LR) and gradually decreasing it. Momentum optimizers are particularly effective at quickly optimizing neural networks within the accelerated schemes category. By revealing the potential interplay between these two types of algorithms (LR and momentum optimizers or momentum rate (MR) in short), in this article, we explore the two variants of SGD algorithms in a single setting. We suggest using cyclic learning as the base optimizer and integrating optimal values of learning rate and momentum rate. The new optimization function proposed in this work is based on the Nesterov accelerated gradient optimizer, which is more efficient computationally and has better generalization capabilities compared to other adaptive optimizers. Results: We investigated the relationship of LR and MR under an important problem of medical image segmentation of cardiac structures from MRI and CT scans. We conducted experiments using the cardiac imaging dataset from the ACDC challenge of MICCAI 2017, and four different architectures shown to be successful for cardiac image segmentation problems. Our comprehensive evaluations demonstrated that the proposed optimizer achieved better results (over a 2% improvement in the dice metric) than other optimizers in deep learning literature with similar or lower computational cost in both single and multi-object segmentation settings. Conclusions: We hypothesized that combination of accelerated and adaptive optimization methods can have a drastic effect in medical image segmentation performances. To this end, we proposed a new cyclic optimization method (CLMR) to address the efficiency and accuracy problems in deep learning based medical image segmentation. The proposed strategy yielded better generalization in comparison to adaptive optimizers.

The Past, Current, and Future of Neonatal Intensive Care Units with Artificial Intelligence

arXiv (Cornell University), Jan 31, 2023

An Efficient Multi-Scale Fusion Network for 3D Organ at Risk (OAR) Segmentation

arXiv (Cornell University), Aug 15, 2022

Accurate segmentation of organs-at-risks (OARs) is a precursor for optimizing radiation therapy p... more Accurate segmentation of organs-at-risks (OARs) is a precursor for optimizing radiation therapy planning. Existing deep learning-based multi-scale fusion architectures have demonstrated a tremendous capacity for 2D medical image segmentation. The key to their success is aggregating global context and maintaining high resolution representations. However, when translated into 3D segmentation problems, existing multi-scale fusion architectures might underperform due to their heavy computation overhead and substantial data diet. To address this issue, we propose a new OAR segmentation framework, called OARFocalFuseNet, which fuses multi-scale features and employs focal modulation for capturing global-local context across multiple scales. Each resolution stream is enriched with features from different resolution scales, and multi-scale information is aggregated to model diverse contextual ranges. As a result, feature representations are further boosted. The comprehensive comparisons in our experimental setup with OAR segmentation as well as multi-organ segmentation show that our proposed OARFocalFuseNet outperforms the recent state-of-theart methods on publicly available OpenKBP datasets and Synapse multi-organ segmentation. Both of the proposed methods (3D-MSF and OARFocalFuseNet) showed promising performance in terms of standard evaluation metrics. Our best performing method (OARFocalFuseNet) obtained a dice coefficient of 0.7995 and hausdorff distance of 5.1435 on OpenKBP datasets and dice coefficient 0.8137 on Synapse multi-organ segmentation dataset.

A Critical Appraisal of Data Augmentation Methods for Imaging-Based Medical Diagnosis Applications

arXiv (Cornell University), Dec 14, 2022

Current data augmentation techniques and transformations are well suited for improving the size a... more Current data augmentation techniques and transformations are well suited for improving the size and quality of natural image datasets but are not yet optimized for medical imaging. We hypothesize that sub-optimal data augmentations can easily distort or occlude medical images, leading to false positives or negatives during patient diagnosis, prediction, or therapy/surgery evaluation. In our experimental results, we found that utilizing commonly used intensity-based data augmentation distorts the MRI scans and leads to texture information loss, thus negatively affecting the overall performance of classification. Additionally, we observed that commonly used data augmentation methods cannot be used with a plug-and-play approach in medical imaging, and requires manual tuning and adjustment.

Microglial iron trafficking: new player in brain injury

Turkish Journal of Medical Sciences

Transformer Based Generative Adversarial Network for Liver Segmentation

Lecture Notes in Computer Science, 2022

Automated liver segmentation from radiology scans (CT, MRI) can improve surgery and therapy plann... more Automated liver segmentation from radiology scans (CT, MRI) can improve surgery and therapy planning and follow-up assessment in addition to conventional use for diagnosis and prognosis. Although convolutional neural networks (CNNs) have became the standard image segmentation tasks, more recently this has started to change towards Transformers based architectures because Transformers are taking advantage of capturing long range dependence modeling capability in signals, so called attention mechanism. In this study, we propose a new segmentation approach using a hybrid approach combining the Transformer(s) with the Generative Adversarial Network (GAN) approach. The premise behind this choice is that the self-attention mechanism of the Transformers allows the network to aggregate the high dimensional feature and provide global information modeling. This mechanism provides better segmentation performance compared with traditional methods. Furthermore, we encode this generator into the GAN based architecture so that the discriminator network in the GAN can classify the credibility of the generated segmentation masks compared with the real masks coming from human (expert) annotations. This allows us to extract the high dimensional topology information in the mask for biomedical image segmentation and provide more reliable segmentation results. Our model achieved a high dice coefficient of 0.9433, recall of 0.9515, and precision of 0.9376 and outperformed other Transformer based approaches. The implementation details of the proposed architecture can be found at https://github.com/UgurDemir/tranformer_liver_segmentation.

Rescaling Creatinine Centiles in Neonates Treated with Therapeutic Hypothermia for Neonatal Encephalopathy

Neonatology

What do Intern Doctors Know about the Use of Pulse Oximetry in Pediatric Patients?

Gazi Medical Journal, 2018

Objective: There is no information on how the knowledge of intern doctors that will work as prima... more Objective: There is no information on how the knowledge of intern doctors that will work as primary health care providers use pulse oximetry in pediatric patients. The aim of this study was to assess and compare knowledge of intern doctors on the use of pulse oximetry especially in pediatric population and factors that affect their knowledge. Methods: A questionnaire consisting of ten questions on the use and principles of pulse oximetry were answered by intern doctors in a university hospital. A uniform answer box including possible response choices was given below each question, and subjects were asked to check the answers that they thought appropriate. Results: A total of 204 questionnaires were analyzed. Only one of third of subjects correctly answered that pulse oximetry measures hypoxia. The known complications of the procedure were not very well known (4-21%). Knowledge about the need for blood gases confirmation in diabetes (19%) and venous congestion (23%) were very low. The correct solutions were answered by 33% of subjects about inappropriate probe position and 26% about intense external light energy. The maximum score of correct answers was 22 in the questionnaire, and mean score of subjects was 12.9 ± 0.17. There was no correlation between age and number of departments with questionnaire scores. Conclusions: The importance of using pulse oximetry should be emphasized. In addition, training about pulse oximetry usage in pediatric patients especially in terms of complications and limitations of devices should be increased in the undergraduate training process.

Musculoskeletal MR Image Segmentation with Artificial Intelligence

Advances in Clinical Radiology

Information Bottleneck Attribution for Visual Explanations of Diagnosis and Prognosis

arXiv (Cornell University), Apr 6, 2021

Visual explanation methods have an important role in the prognosis of the patients where the anno... more Visual explanation methods have an important role in the prognosis of the patients where the annotated data is limited or unavailable. There have been several attempts to use gradient-based attribution methods to localize pathology from medical scans without using segmentation labels. This research direction has been impeded by the lack of robustness and reliability. These methods are highly sensitive to the network parameters. In this study, we introduce a robust visual explanation method to address this problem for medical applications. We provide an innovative visual explanation algorithm for general purpose and as an example application we demonstrate its effectiveness for quantifying lesions in the lungs caused by the Covid-19 with high accuracy and robustness without using dense segmentation labels. This approach overcomes the drawbacks of commonly used Grad-CAM and its extended versions. The premise behind our proposed strategy is that the information flow is minimized while ensuring the classifier prediction stays similar. Our findings indicate that the bottleneck condition provides a more stable severity estimation than the similar attribution methods. The source code will be publicly available upon publication.

Selecting the best optimizers for deep learning–based medical image segmentation

Frontiers in Radiology

PurposeThe goal of this work is to explore the best optimizers for deep learning in the context o... more PurposeThe goal of this work is to explore the best optimizers for deep learning in the context of medical image segmentation and to provide guidance on how to design segmentation networks with effective optimization strategies.ApproachMost successful deep learning networks are trained using two types of stochastic gradient descent (SGD) algorithms: adaptive learning and accelerated schemes. Adaptive learning helps with fast convergence by starting with a larger learning rate (LR) and gradually decreasing it. Momentum optimizers are particularly effective at quickly optimizing neural networks within the accelerated schemes category. By revealing the potential interplay between these two types of algorithms [LR and momentum optimizers or momentum rate (MR) in short], in this article, we explore the two variants of SGD algorithms in a single setting. We suggest using cyclic learning as the base optimizer and integrating optimal values of learning rate and momentum rate. The new optimi...

A review of deep learning and radiomics approaches for pancreatic cancer diagnosis from medical imaging

Current Opinion in Gastroenterology

Purpose of review Early and accurate diagnosis of pancreatic cancer is crucial for improving pati... more Purpose of review Early and accurate diagnosis of pancreatic cancer is crucial for improving patient outcomes, and artificial intelligence (AI) algorithms have the potential to play a vital role in computer-aided diagnosis of pancreatic cancer. In this review, we aim to provide the latest and relevant advances in AI, specifically deep learning (DL) and radiomics approaches, for pancreatic cancer diagnosis using cross-sectional imaging examinations such as computed tomography (CT) and magnetic resonance imaging (MRI). Recent findings This review highlights the recent developments in DL techniques applied to medical imaging, including convolutional neural networks (CNNs), transformer-based models, and novel deep learning architectures that focus on multitype pancreatic lesions, multiorgan and multitumor segmentation, as well as incorporating auxiliary information. We also discuss advancements in radiomics, such as improved imaging feature extraction, optimized machine learning classif...

Tu1992 AI EMPOWERED AUTOMATIC VOLUME DELINEATION OF LIVER FROM CT SCANS FOR DIAGNOSTIC WORKFLOW

Gastroenterology

Tu1986 A FULLY AUTOMATIC AI SYSTEM FOR PANCREAS SEGMENTATION FROM MULTICENTER MRI SCANS

Gastroenterology

Ensuring Trustworthy Medical Artificial Intelligencethrough Ethical and Philosophical Principles

arXiv (Cornell University), Apr 23, 2023

Artificial Intelligence (AI) methods hold immense potential to revolutionize numerous medical car... more Artificial Intelligence (AI) methods hold immense potential to revolutionize numerous medical care by enhancing the experience of medical experts and patients. AI-based computer-assisted diagnosis and treatment tools can democratize healthcare by matching the clinical level or surpassing clinical experts. As a result, advanced healthcare services can be affordable to all populations, irrespective of demographics, race, or socioeconomic background. The democratization of such AI tools can reduce the cost of care, optimize resource allocation, and improve the quality of care. In contrast to humans, AI can uncover complex relations in the data from a large set of inputs and even lead to new evidence-based knowledge in medicine. However, integrating AI into healthcare raises several ethical and philosophical concerns, such as bias, transparency, autonomy, responsibility, and accountability. Here, we emphasize recent advances in AI-assisted medical image analysis, existing standards, and the significance of comprehending ethical issues and best practices for clinical settings. We cover the technical and ethical challenges and implications of deploying AI in hospitals and public organizations. We also discuss key measures and techniques to address ethical challenges, data scarcity, racial bias, lack of transparency, and algorithmic bias and provide recommendations and future directions.

Building Blocks of AI

AI in Clinical Medicine, Mar 3, 2023

Asphyxia Associated Metabolite Biomarker Investigation (AAMBI). Ergebnisse im Neugeborenenalter und mit 22-42 Monaten

30. Kongress der Deutschen Gesellschaft für Perinatale Medizin – „Wandel als Herausforderung“, 2021

Mortality from gastrointestinal congenital anomalies at 264 hospitals in 74 low-income, middle-income, and high-income countries: a multicentre, international, prospective cohort study

The Lancet, 2021

Background Congenital anomalies are the fifth leading cause of mortality in children younger than... more Background Congenital anomalies are the fifth leading cause of mortality in children younger than 5 years globally. Many gastrointestinal congenital anomalies are fatal without timely access to neonatal surgical care, but few studies have been done on these conditions in low-income and middle-income countries (LMICs). We compared outcomes of the seven most common gastrointestinal congenital anomalies in low-income, middle-income, and high-income countries globally, and identified factors associated with mortality. Methods We did a multicentre, international prospective cohort study of patients younger than 16 years, presenting to hospital for the first time with oesophageal atresia, congenital diaphragmatic hernia, intestinal atresia, gastroschisis, exomphalos, anorectal malformation, and Hirschsprung's disease. Recruitment was of consecutive patients for a minimum of 1 month between October, 2018, and April, 2019. We collected data on patient demographics, clinical status, interventions, and outcomes using the REDCap platform. Patients were followed up for 30 days after primary intervention, or 30 days after admission if they did not receive an intervention. The primary outcome was all-cause, in-hospital mortality for all conditions combined and each condition individually, stratified by country income status. We did a complete case analysis. Findings We included 3849 patients with 3975 study conditions (560 with oesophageal atresia, 448 with congenital diaphragmatic hernia, 681 with intestinal atresia, 453 with gastroschisis, 325 with exomphalos, 991 with anorectal malformation, and 517 with Hirschsprung's disease) from 264 hospitals (89 in high-income countries, 166 in middleincome countries, and nine in low-income countries) in 74 countries. Of the 3849 patients, 2231 (58•0%) were male. Median gestational age at birth was 38 weeks (IQR 36-39) and median bodyweight at presentation was 2•8 kg (2•3-3•3). Mortality among all patients was 37 (39•8%) of 93 in low-income countries, 583 (20•4%) of 2860 in middle-income countries, and 50 (5•6%) of 896 in high-income countries (p<0•0001 between all country income groups). Gastroschisis had the greatest difference in mortality between country income strata (nine [90•0%] of ten in lowincome countries, 97 [31•9%] of 304 in middle-income countries, and two [1•4%] of 139 in high-income countries; p≤0•0001 between all country income groups). Factors significantly associated with higher mortality for all patients combined included country income status (low-income vs high-income countries, risk ratio 2•78 [95% CI 1•88-4•11], p<0•0001; middle-income vs high-income countries, 2•11 [1•59-2•79], p<0•0001), sepsis at presentation (1•20 [1•04-1•40], p=0•016), higher American Society of Anesthesiologists (ASA) score at primary intervention (ASA 4-5 vs ASA 1-2, 1•82 [1•40-2•35], p<0•0001; ASA 3 vs ASA 1-2, 1•58, [1•30-1•92], p<0•0001]), surgical safety checklist not used (1•39 [1•02-1•90], p=0•035), and ventilation or parenteral nutrition unavailable when needed (ventilation 1•96, [1•41-2•71], p=0•0001; parenteral nutrition 1•35, [1•05-1•74], p=0•018). Administration of parenteral nutrition (0•61, [0•47-0•79], p=0•0002) and use of a peripherally inserted central catheter (0•65 [0•50-0•86], p=0•0024) or percutaneous central line (0•69 [0•48-1•00], p=0•049) were associated with lower mortality. Interpretation Unacceptable differences in mortality exist for gastrointestinal congenital anomalies between lowincome, middle-income, and high-income countries. Improving access to quality neonatal surgical care in LMICs will be vital to achieve Sustainable Development Goal 3.2 of ending preventable deaths in neonates and children younger than 5 years by 2030. Funding Wellcome Trust.

Die AAMBI-Studie (Asphyxia Associated Metabolite Biomarker Investigation): Ergebnisse im Neugeborenenalter und mit 22–42 Monaten

Zeitschrift für Geburtshilfe und Neonatologie

Deep Learning Algorithms for Pancreas Segmentation from Radiology Scans

Advances in Clinical Radiology

Selecting the Best Optimizers for Deep Learning based Medical Image Segmentation

arXiv (Cornell University), Feb 4, 2023

The goal of this work is to identify the best optimizers for deep learning in the context of card... more The goal of this work is to identify the best optimizers for deep learning in the context of cardiac image segmentation and to provide guidance on how to design segmentation networks with effective optimization strategies. Approach: Most successful deep learning networks are trained using two types of stochastic gradient descent (SGD) algorithms: adaptive learning and accelerated schemes. Adaptive learning helps with fast convergence by starting with a larger learning rate (LR) and gradually decreasing it. Momentum optimizers are particularly effective at quickly optimizing neural networks within the accelerated schemes category. By revealing the potential interplay between these two types of algorithms (LR and momentum optimizers or momentum rate (MR) in short), in this article, we explore the two variants of SGD algorithms in a single setting. We suggest using cyclic learning as the base optimizer and integrating optimal values of learning rate and momentum rate. The new optimization function proposed in this work is based on the Nesterov accelerated gradient optimizer, which is more efficient computationally and has better generalization capabilities compared to other adaptive optimizers. Results: We investigated the relationship of LR and MR under an important problem of medical image segmentation of cardiac structures from MRI and CT scans. We conducted experiments using the cardiac imaging dataset from the ACDC challenge of MICCAI 2017, and four different architectures shown to be successful for cardiac image segmentation problems. Our comprehensive evaluations demonstrated that the proposed optimizer achieved better results (over a 2% improvement in the dice metric) than other optimizers in deep learning literature with similar or lower computational cost in both single and multi-object segmentation settings. Conclusions: We hypothesized that combination of accelerated and adaptive optimization methods can have a drastic effect in medical image segmentation performances. To this end, we proposed a new cyclic optimization method (CLMR) to address the efficiency and accuracy problems in deep learning based medical image segmentation. The proposed strategy yielded better generalization in comparison to adaptive optimizers.

The Past, Current, and Future of Neonatal Intensive Care Units with Artificial Intelligence

arXiv (Cornell University), Jan 31, 2023

An Efficient Multi-Scale Fusion Network for 3D Organ at Risk (OAR) Segmentation

arXiv (Cornell University), Aug 15, 2022

Accurate segmentation of organs-at-risks (OARs) is a precursor for optimizing radiation therapy p... more Accurate segmentation of organs-at-risks (OARs) is a precursor for optimizing radiation therapy planning. Existing deep learning-based multi-scale fusion architectures have demonstrated a tremendous capacity for 2D medical image segmentation. The key to their success is aggregating global context and maintaining high resolution representations. However, when translated into 3D segmentation problems, existing multi-scale fusion architectures might underperform due to their heavy computation overhead and substantial data diet. To address this issue, we propose a new OAR segmentation framework, called OARFocalFuseNet, which fuses multi-scale features and employs focal modulation for capturing global-local context across multiple scales. Each resolution stream is enriched with features from different resolution scales, and multi-scale information is aggregated to model diverse contextual ranges. As a result, feature representations are further boosted. The comprehensive comparisons in our experimental setup with OAR segmentation as well as multi-organ segmentation show that our proposed OARFocalFuseNet outperforms the recent state-of-theart methods on publicly available OpenKBP datasets and Synapse multi-organ segmentation. Both of the proposed methods (3D-MSF and OARFocalFuseNet) showed promising performance in terms of standard evaluation metrics. Our best performing method (OARFocalFuseNet) obtained a dice coefficient of 0.7995 and hausdorff distance of 5.1435 on OpenKBP datasets and dice coefficient 0.8137 on Synapse multi-organ segmentation dataset.

A Critical Appraisal of Data Augmentation Methods for Imaging-Based Medical Diagnosis Applications

arXiv (Cornell University), Dec 14, 2022

Current data augmentation techniques and transformations are well suited for improving the size a... more Current data augmentation techniques and transformations are well suited for improving the size and quality of natural image datasets but are not yet optimized for medical imaging. We hypothesize that sub-optimal data augmentations can easily distort or occlude medical images, leading to false positives or negatives during patient diagnosis, prediction, or therapy/surgery evaluation. In our experimental results, we found that utilizing commonly used intensity-based data augmentation distorts the MRI scans and leads to texture information loss, thus negatively affecting the overall performance of classification. Additionally, we observed that commonly used data augmentation methods cannot be used with a plug-and-play approach in medical imaging, and requires manual tuning and adjustment.

Microglial iron trafficking: new player in brain injury

Turkish Journal of Medical Sciences

Transformer Based Generative Adversarial Network for Liver Segmentation

Lecture Notes in Computer Science, 2022

Automated liver segmentation from radiology scans (CT, MRI) can improve surgery and therapy plann... more Automated liver segmentation from radiology scans (CT, MRI) can improve surgery and therapy planning and follow-up assessment in addition to conventional use for diagnosis and prognosis. Although convolutional neural networks (CNNs) have became the standard image segmentation tasks, more recently this has started to change towards Transformers based architectures because Transformers are taking advantage of capturing long range dependence modeling capability in signals, so called attention mechanism. In this study, we propose a new segmentation approach using a hybrid approach combining the Transformer(s) with the Generative Adversarial Network (GAN) approach. The premise behind this choice is that the self-attention mechanism of the Transformers allows the network to aggregate the high dimensional feature and provide global information modeling. This mechanism provides better segmentation performance compared with traditional methods. Furthermore, we encode this generator into the GAN based architecture so that the discriminator network in the GAN can classify the credibility of the generated segmentation masks compared with the real masks coming from human (expert) annotations. This allows us to extract the high dimensional topology information in the mask for biomedical image segmentation and provide more reliable segmentation results. Our model achieved a high dice coefficient of 0.9433, recall of 0.9515, and precision of 0.9376 and outperformed other Transformer based approaches. The implementation details of the proposed architecture can be found at https://github.com/UgurDemir/tranformer_liver_segmentation.

Rescaling Creatinine Centiles in Neonates Treated with Therapeutic Hypothermia for Neonatal Encephalopathy

Neonatology

What do Intern Doctors Know about the Use of Pulse Oximetry in Pediatric Patients?

Gazi Medical Journal, 2018

Objective: There is no information on how the knowledge of intern doctors that will work as prima... more Objective: There is no information on how the knowledge of intern doctors that will work as primary health care providers use pulse oximetry in pediatric patients. The aim of this study was to assess and compare knowledge of intern doctors on the use of pulse oximetry especially in pediatric population and factors that affect their knowledge. Methods: A questionnaire consisting of ten questions on the use and principles of pulse oximetry were answered by intern doctors in a university hospital. A uniform answer box including possible response choices was given below each question, and subjects were asked to check the answers that they thought appropriate. Results: A total of 204 questionnaires were analyzed. Only one of third of subjects correctly answered that pulse oximetry measures hypoxia. The known complications of the procedure were not very well known (4-21%). Knowledge about the need for blood gases confirmation in diabetes (19%) and venous congestion (23%) were very low. The correct solutions were answered by 33% of subjects about inappropriate probe position and 26% about intense external light energy. The maximum score of correct answers was 22 in the questionnaire, and mean score of subjects was 12.9 ± 0.17. There was no correlation between age and number of departments with questionnaire scores. Conclusions: The importance of using pulse oximetry should be emphasized. In addition, training about pulse oximetry usage in pediatric patients especially in terms of complications and limitations of devices should be increased in the undergraduate training process.

Musculoskeletal MR Image Segmentation with Artificial Intelligence

Advances in Clinical Radiology