Hasil untuk "Computer applications to medicine. Medical informatics"

Jamil Zaghir, M. Naguib, Mina Bjelogrlic et al.

Background Prompt engineering, focusing on crafting effective prompts to large language models (LLMs), has garnered attention for its capabilities at harnessing the potential of LLMs. This is even more crucial in the medical domain due to its specialized terminology and language technicity. Clinical natural language processing applications must navigate complex language and ensure privacy compliance. Prompt engineering offers a novel approach by designing tailored prompts to guide models in exploiting clinically relevant information from complex medical texts. Despite its promise, the efficacy of prompt engineering in the medical domain remains to be fully explored. Objective The aim of the study is to review research efforts and technical approaches in prompt engineering for medical applications as well as provide an overview of opportunities and challenges for clinical practice. Methods Databases indexing the fields of medicine, computer science, and medical informatics were queried in order to identify relevant published papers. Since prompt engineering is an emerging field, preprint databases were also considered. Multiple data were extracted, such as the prompt paradigm, the involved LLMs, the languages of the study, the domain of the topic, the baselines, and several learning, design, and architecture strategies specific to prompt engineering. We include studies that apply prompt engineering–based methods to the medical domain, published between 2022 and 2024, and covering multiple prompt paradigms such as prompt learning (PL), prompt tuning (PT), and prompt design (PD). Results We included 114 recent prompt engineering studies. Among the 3 prompt paradigms, we have observed that PD is the most prevalent (78 papers). In 12 papers, PD, PL, and PT terms were used interchangeably. While ChatGPT is the most commonly used LLM, we have identified 7 studies using this LLM on a sensitive clinical data set. Chain-of-thought, present in 17 studies, emerges as the most frequent PD technique. While PL and PT papers typically provide a baseline for evaluating prompt-based approaches, 61% (48/78) of the PD studies do not report any nonprompt-related baseline. Finally, we individually examine each of the key prompt engineering–specific information reported across papers and find that many studies neglect to explicitly mention them, posing a challenge for advancing prompt engineering research. Conclusions In addition to reporting on trends and the scientific landscape of prompt engineering, we provide reporting guidelines for future studies to help advance research in the medical field. We also disclose tables and figures summarizing medical prompt engineering papers available and hope that future contributions will leverage these existing works to better advance the field.

Marcos Gabriel Mendes Lauande, Geraldo Braz Júnior, João Dallyson Sousa de Almeida et al.

Penile cancer has an incidence strongly linked to sociocultural factors, being more common in underdeveloped countries like Brazil, where it represents approximately 2% of cancers affecting men. This dataset was created to address the scarcity of publicly available resources for classifying histopathological images in penile cancer research. The images were collected in 2021 from tissue samples obtained through biopsies of patients undergoing treatment for penile cancer. After staining with Hematoxylin and Eosin (H&E), the tissue samples were photographed using a Leica ICC50 HD camera attached to a bright-field microscope (Leica DM500). The dataset comprises 194 high-resolution images (2048 × 1536 pixels), categorized by magnification (40X and 100X) and pathological classification (Tumor or Non-Tumor). Metadata includes additional information such as histological grade and, for some images, HPV status. Although previous works have focused primarily on binary classification tasks, the dataset includes additional labels, such as histological grade and HPV (Human Papilloma Virus) presence, which provide opportunities for multi-label classification or other types of predictive modelling. These extended labels enhance the dataset’s versatility for more complex tasks in medical image analysis. The dataset holds significant reuse potential for machine learning tasks beyond binary classification, allowing researchers to explore additional layers of analysis, such as HPV detection and histological grading. It can also be used for model benchmarking and comparative studies in cancer research, contributing to developing new diagnostic tools. The dataset and metadata are available for further research and model development.

Irena Roterman-Konieczna

This editorial article outlines the origins, development and scientific mission of Bio-Algorithms and Med-Systems on the occasion of its 20th anniversary. It reconstructs the historical context of early 21st-century Poland, when interdisciplinary collaboration between medicine, computer science and engineering was still uncommon and often meet with scepticism. The text describes the pioneering role of the Jagiellonian University Medical College and AGH University of Krakow in promoting biomedical informatics, cybernetics and biomedical engineering fields that would later become essential to modern healthcare. It also recounts the establishment of the journal in 2005 as a response to the lack of publication venues for interdisciplinary work combining bio-phenomena, technical sciences and medical applications. The article presents the journal’s contribution to shaping the newly emerging discipline of biomedical engineering in Poland, its early publishing philosophy, and its evolution through various editorial and publishing stages. Finally, the authors reflect on the journal’s legacy, emphasising the importance of interdisciplinary cooperation, technological innovation and ethical frameworks as prerequisites for scientific progress.

Muath Aldergham, Areeg Alfouri, Rasha Al Madat

Artificial intelligence in medicine refers to the use of machine learning models to help process medical data and provide medical professionals with important insights, improving health outcomes and patient experience. Thanks to recent advances in computer science and informatics, artificial intelligence (AI) is rapidly becoming an integral part of modern healthcare. Therefore, artificial intelligence algorithms and other AI-powered applications are now used to support medical professionals in clinical settings and in ongoing research. There are several applications of Artificial intelligence in medicine, including applications to help detect and diagnose diseases; applications to treat diseases with the help of an AI-powered virtual assistant; AI applications in medical imaging; applications to increase the efficiency of clinical trials; and applications to accelerate drug development. The benefits of Artificial intelligence in medicine can be summarized in providing informed patient care, reducing errors, reducing care costs, and increasing doctor-patient engagement.

Marek Klimo, M. Kvaššay, N. Kvassayová

Three-dimensional (3D) models have significantly transformed medical practice, education, and research in the healthcare field. This study examines the diverse applications of 3D models in the field of medicine, encompassing a wide range of applications such as basic organ simulations, intricate surgical procedures, and personalized medicine. These models provide a comprehensive depiction of anatomical features and clinical settings using sophisticated imaging techniques and Computer-Aided Design (CAD) tools. Furthermore, the integration of Virtual Reality (VR) and Augmented Reality (AR) technology has significantly improved the usefulness of 3D models, offering immersive experiences and innovative opportunities for studying and executing complex tasks. However, one of the key issues in the development of all these applications connecting informatics and medicine is the quality of 3D models of human anatomy. In this paper, we compare two generic male models, analyze them quantitatively and introduce one more model that takes the best of both. We then compare newly modified model and quantitatively compare it with the previous models. In the development of the model, we focus on human male skeletal model.

Ashagrie Sharew Iyassu, Haile Mekonnen Fenta, Zelalem G. Dessie et al.

Abstract Background In causal analyses, some third factor may distort the relationship between the exposure and the outcome variables under study, which gives spurious results. In this case, treatment groups and control groups that receive and do not receive the exposure are different from one another in some other essential variables, called confounders. Method Place of birth was used as exposure variable and age-specific childhood vaccination status was used as outcome variables. Three approaches of confounder selection techniques such as all pre-treatment covariates, outcome cause covariates, and common cause covariates were proposed. Multiple logistic regression was used to estimate the propensity score for inverse probability treatment weighting (IPTW) confounder adjustment techniques. The proportional odds model was used to estimate the causal effect of place of birth on age-specific childhood vaccination. To validate the result obtained from observed data, we used a plasmode simulation of resampling 1000 samples from actual data 500 times. Result Outcome cause and common cause confounder identification techniques gave comparable results in terms of treatment effect in the plasmode data. However, outcome causes that contain common causes and predictors of the outcome confounder identification gave relatively better treatment effect results. The treatment effect result in the IPTW confounder adjustment method was better than that of the regression adjustment method. The effect of place of birth on log odds of cumulative probability of age-specific childhood vaccination was 0.36 with odds ratio of 1.43 for higher level vaccination status. Conclusion It is essential to use plasmode simulation data to validate the reproducibility of the proposed methods on the observed data. It is important to use outcome-cause covariates to adjust their confounding effect on the outcome. Using inverse probability treatment weighting gives unbiased treatment effect results as compared to the regression method of confounder adjustment. Institutional delivery increases the likelihood of childhood vaccination at the recommended schedule.

Ahmed Medhat Zayed, Arne Janssens, Pavlos Mamouris et al.

Abstract Background The integrity of clinical research and machine learning models in healthcare heavily relies on the quality of underlying clinical laboratory data. However, the preprocessing of this data to ensure its reliability and accuracy remains a significant challenge due to variations in data recording and reporting standards. Methods We developed lab2clean, a novel algorithm aimed at automating and standardizing the cleaning of retrospective clinical laboratory results data. lab2clean was implemented as two R functions specifically designed to enhance data conformance and plausibility by standardizing result formats and validating result values. The functionality and performance of the algorithm were evaluated using two extensive electronic medical record (EMR) databases, encompassing various clinical settings. Results lab2clean effectively reduced the variability of laboratory results and identified potentially erroneous records. Upon deployment, it demonstrated effective and fast standardization and validation of substantial laboratory data records. The evaluation highlighted significant improvements in the conformance and plausibility of lab results, confirming the algorithm’s efficacy in handling large-scale data sets. Conclusions lab2clean addresses the challenge of preprocessing and cleaning clinical laboratory data, a critical step in ensuring high-quality data for research outcomes. It offers a straightforward, efficient tool for researchers, improving the quality of clinical laboratory data, a major portion of healthcare data. Thereby, enhancing the reliability and reproducibility of clinical research outcomes and clinical machine learning models. Future developments aim to broaden its functionality and accessibility, solidifying its vital role in healthcare data management. Graphical Abstract

Quoc Cuong Ngo, Nicole McConnell, Mohammod Abdul Motin et al.

Objective: A change in handwriting is an early sign of Parkinson’s disease (PD). However, significant inter-person differences in handwriting make it difficult to identify pathological handwriting, especially in the early stages. This paper reports the testing of NeuroDiag, a software-based medical device, for the automated detection of PD using handwriting patterns. NeuroDiag is designed to direct the user to perform six drawing and writing tasks, and the recordings are then uploaded onto a server for analysis. Kinematic information and pen pressure of handwriting are extracted and used as baseline parameters. NeuroDiag was trained based on 26 PD patients in the early stage of the disease and 26 matching controls. Methods: Twenty-three people with PD (PPD) in their early stage of the disease, 25 age-matched healthy controls (AMC), and 7 young healthy controls were recruited for this study. Under the supervision of a consultant neurologist or their nurse, the participants used NeuroDiag. The reports were generated in real-time and tabulated by an independent observer. Results: The participants were able to use NeuroDiag without assistance. The handwriting data was successfully uploaded to the server where the report was automatically generated in real-time. There were significant differences in the writing speed between PPD and AMC (P<0.001). NeuroDiag showed 86.96% sensitivity and 76.92% specificity in differentiating PPD from those without PD. Conclusion: In this work, we tested the reliability of NeuroDiag in differentiating between PPD and AMC for real-time applications. The results show that NeuroDiag has the potential to be used to assist neurologists and for telehealth applications. Clinical and Translational Impact Statement — This pre-clinical study shows the feasibility of developing a community-wide screening program for Parkinson’s disease using automated handwriting analysis software, NeuroDiag.

Ajay Dadhich, Jaideep Patel, Rovin Tiwari et al.

Mind-wandering (MW) is when an individual’s concentration drifts away from the task or activity. Researchers found a greater variability in electroencephalogram (EEG) signals due to MW. Collecting more nuanced information from raw EEG data to examine the harmful effects of MW is time-consuming. This study proposes a multi-resolution assessment of EEG signals using the flexible analytic wavelet transform (FAWT). The FAWT algorithm decomposes raw EEG data into more representative sub-bands (SBs). Several statistical characteristics are derived from the obtained SBs, and the effects of MW during meditation on the EEG signals are investigated. A set of significant characteristics is chosen and fed into the machine learning modules using a 10-fold validation approach to detect MW subjects automatically. Our proposed framework attained the highest classification accuracy of 92.41%, the highest sensitivity of 93.56%, and the highest specificity of 91.97%. The proposed framework can be used to design a suitable brain-computer interface (BCI) system to reduce MW and increase meditation depth for holistic and long-term health in society.

Ligita Mažeikė, Arminas Vareika

Tinkama miego trukmė ir kokybė yra būtini optimaliai psichinei ir fizinei sveikatai. Karinės profesijos susiduria su unikaliais iššūkiais, tokiais kaip 36 valandų darbo pamainos, fiziškai alinantis darbas ir situacijos, dėl kurių galima susižaloti arba žūti. Pervargę kariai kenčia nuo sumažėjusio budrumo, su sprendimais susijusio reakcijos laiko, trumpalaikės atminties, navigacijos įgūdžių ir, kai kuriais atvejais, taiklumo per treniruotes suprastėjimo. Didelio fizinio ir psichinio nuovargio derinys gali padidinti traumų riziką ir sumažinti gebėjimą priimti tinkamus sprendimus reikiamu laiku. Šio tyrimo tikslas – nustatyti karinių pratybų poveikį profesionalių karių miego kokybei, kadangi dažniausi miego nepakankamumo atvejai pasitaiko per karines pratybas. Į tyrimą buvo įtraukta 10 profesionalių Lietuvos kariuomenės žvalgų būrio karių, 32,8 ± 6,9 metų amžiaus, atitinkančių pačius aukščiausius karių fizinio parengimo testo reikalavimus (≥ 270 balų). Profesionalių karių miego kokybė buvo vertinama 7 paras prieš karines pratybas, 7 paras karinių pratybų metu ir 9 paras iš karto po karinių pratybų. Visi tiriamieji dėvėjo laikrodžius „Garmin Decent G1“, kuriais buvo fiksuojami miego parodymai. Buvo registruojama bendra miego trukmė, gilaus miego, lengvo miego, REM ir būdravimo fazių trukmė. Nustatyta, kad karinių pratybų metu, karių bendra miego trukmė buvo 5,3 ± 2,9 val. per dieną (toliau – val./d.), o prieš pratybas buvo 2,3 ± 1,1 val./d. ilgesnė (p < 0,05). Po karinių pratybų karių bendra miego trukmė buvo 3,0 ± 1,1 val./d. ilgesnė nei karinių pratybų metu (p < 0,05). Taip pat po karinių pratybų karių bendra miego trukmė buvo ilgesnė 0,7 ± 0,5 val./d. lyginat su miego trukme prieš pratybas (p < 0,05). Karinių pratybų metu taip pat sutrumpėjo gilaus miego fazė. Gilaus miego fazės trukmė pratybų metu sutrumpėjo net 58 proc. (p < 0,05) lyginant su prieš pratybas buvusia, tačiau po pratybų išliko 18 proc. trumpesnė nei prieš pratybas (p < 0,05). Lengvo miego fazės trukmė buvo 93,7 ± 46,6 min. per dieną (toliau – min./d.) trumpesnė nei prieš pratybas (p < 0,05), tačiau vertinant procentais lengvo miego fazė sudarė apie 60 proc. bendros miego trukmės visuose etapuose ir tik apie 2 proc. buvo trumpesnė po karinių pratybų (p > 0,05). Karinių pratybų metu REM ir būdravimo miego fazių trukmė reikšmingai nesiskyrė nuo prieš pratybas buvusių trukmių (p > 0,05).

Nicholas Konz, Richard Osuala, Preeti Verma et al.

Determining whether two sets of images belong to the same or different distributions or domains is a crucial task in modern medical image analysis and deep learning; for example, to evaluate the output quality of image generative models. Currently, metrics used for this task either rely on the (potentially biased) choice of some downstream task, such as segmentation, or adopt task-independent perceptual metrics (e.g., Fréchet Inception Distance/FID) from natural imaging, which we show insufficiently capture anatomical features. To this end, we introduce a new perceptual metric tailored for medical images, FRD (Fréchet Radiomic Distance), which utilizes standardized, clinically meaningful, and interpretable image features. We show that FRD is superior to other image distribution metrics for a range of medical imaging applications, including out-of-domain (OOD) detection, the evaluation of image-to-image translation (by correlating more with downstream task performance as well as anatomical consistency and realism), and the evaluation of unconditional image generation. Moreover, FRD offers additional benefits such as stability and computational efficiency at low sample sizes, sensitivity to image corruptions and adversarial attacks, feature interpretability, and correlation with radiologist-perceived image quality. Additionally, we address key gaps in the literature by presenting an extensive framework for the multifaceted evaluation of image similarity metrics in medical imaging -- including the first large-scale comparative study of generative models for medical image translation -- and release an accessible codebase to facilitate future research. Our results are supported by thorough experiments spanning a variety of datasets, modalities, and downstream tasks, highlighting the broad potential of FRD for medical image analysis.

Frederik F. Flöther

The last few years have seen rapid progress in transitioning quantum computing from lab to industry. In healthcare and life sciences, more than 40 proof-of-concept experiments and studies have been conducted; an increasing number of these are even run on real quantum hardware. Major investments have been made with hundreds of millions of dollars already allocated towards quantum applications and hardware in medicine. In addition to pharmaceutical and life sciences uses, clinical and medical applications are now increasingly coming into the picture. This chapter focuses on three key use case areas associated with (precision) medicine, including genomics and clinical research, diagnostics, and treatments and interventions. Examples of organizations and the use cases they have been researching are given; ideas how the development of practical quantum computing applications can be further accelerated are described.

Marion Hamard, Marta Sans Merce, Karel Gorican et al.

Osteoarthritis (OA) is a prevalent disease and the leading cause of pain, disability, and quality of life deterioration. Our study sought to evaluate the image quality and dose of cone-beam computed tomography arthrography (CBCT-A) and compare them to digital radiography (DR) for OA diagnoses. Overall, 32 cases of CBCT-A and DR with OA met the inclusion criteria and were prospectively analyzed. The Kellgren and Lawrence classification (KLC) stage, sclerosis, osteophytes, erosions, and mean joint width (MJW) were compared between CBCT-A and DR. Image quality was excellent in all CBCT-A cases, with excellent inter-observer agreement. OA under-classification was noticed with DR for MJW (<i>p</i> = 0.02), osteophyte detection (<0.0001), and KLC (<i>p</i> < 0.0001). The Hounsfield Unit (HU) values obtained for the cone-beam computed tomography CBCT did not correspond to the values for multi-detector computed tomography (MDCT), with a greater mean deviation obtained with the MDCT HU for Modeled Based Iterative Reconstruction 1st (MBIR1) than for the 2nd generation (MBIR2). CBCT-A has been found to be more reliable for OA diagnosis than DR as revealed by our results using a three-point rating scale for the qualitative image analysis, with higher quality and an acceptable dose. Moreover, the use of this imaging technique permits the preoperative assessment of extremities in an OA diagnosis, with the upright position and bone microarchitecture analysis being two other advantages of CBCT-A.

Md. Zihad Bin Jahangir, Ruksat Hossain, Riadul Islam et al.

Medical imaging informatics is a rapidly growing field that combines the principles of medical imaging and informatics to improve the acquisition, management, and interpretation of medical images. This chapter introduces the basic concepts of medical imaging informatics, including image processing, feature engineering, and machine learning. It also discusses the recent advancements in computer vision and deep learning technologies and how they are used to develop new quantitative image markers and prediction models for disease detection, diagnosis, and prognosis prediction. By covering the basic knowledge of medical imaging informatics, this chapter provides a foundation for understanding the role of informatics in medicine and its potential impact on patient care.

Emma A. M. Stanley, Raissa Souza, Anthony Winder et al.

Artificial intelligence (AI) models trained using medical images for clinical tasks often exhibit bias in the form of disparities in performance between subgroups. Since not all sources of biases in real-world medical imaging data are easily identifiable, it is challenging to comprehensively assess how those biases are encoded in models, and how capable bias mitigation methods are at ameliorating performance disparities. In this article, we introduce a novel analysis framework for systematically and objectively investigating the impact of biases in medical images on AI models. We developed and tested this framework for conducting controlled in silico trials to assess bias in medical imaging AI using a tool for generating synthetic magnetic resonance images with known disease effects and sources of bias. The feasibility is showcased by using three counterfactual bias scenarios to measure the impact of simulated bias effects on a convolutional neural network (CNN) classifier and the efficacy of three bias mitigation strategies. The analysis revealed that the simulated biases resulted in expected subgroup performance disparities when the CNN was trained on the synthetic datasets. Moreover, reweighing was identified as the most successful bias mitigation strategy for this setup, and we demonstrated how explainable AI methods can aid in investigating the manifestation of bias in the model using this framework. Developing fair AI models is a considerable challenge given that many and often unknown sources of biases can be present in medical imaging datasets. In this work, we present a novel methodology to objectively study the impact of biases and mitigation strategies on deep learning pipelines, which can support the development of clinical AI that is robust and responsible.

Frederik F. Flöther

Medicine, including fields in healthcare and life sciences, has seen a flurry of quantum-related activities and experiments in the last few years (although biology and quantum theory have arguably been entangled ever since Schrödinger's cat). The initial focus was on biochemical and computational biology problems; recently, however, clinical and medical quantum solutions have drawn increasing interest. The rapid emergence of quantum computing in health and medicine necessitates a mapping of the landscape. In this review, clinical and medical proof-of-concept quantum computing applications are outlined and put into perspective. These consist of over 40 experimental and theoretical studies. The use case areas span genomics, clinical research and discovery, diagnostics, and treatments and interventions. Quantum machine learning (QML) in particular has rapidly evolved and shown to be competitive with classical benchmarks in recent medical research. Near-term QML algorithms have been trained with diverse clinical and real-world data sets. This includes studies in generating new molecular entities as drug candidates, diagnosing based on medical image classification, predicting patient persistence, forecasting treatment effectiveness, and tailoring radiotherapy. The use cases and algorithms are summarized and an outlook on medicine in the quantum era, including technical and ethical challenges, is provided.

Jianing Qiu, Lin Li, Jiankai Sun et al.

Large AI models, or foundation models, are models recently emerging with massive scales both parameter-wise and data-wise, the magnitudes of which can reach beyond billions. Once pretrained, large AI models demonstrate impressive performance in various downstream tasks. A prime example is ChatGPT, whose capability has compelled people's imagination about the far-reaching influence that large AI models can have and their potential to transform different domains of our lives. In health informatics, the advent of large AI models has brought new paradigms for the design of methodologies. The scale of multi-modal data in the biomedical and health domain has been ever-expanding especially since the community embraced the era of deep learning, which provides the ground to develop, validate, and advance large AI models for breakthroughs in health-related areas. This article presents a comprehensive review of large AI models, from background to their applications. We identify seven key sectors in which large AI models are applicable and might have substantial influence, including 1) bioinformatics; 2) medical diagnosis; 3) medical imaging; 4) medical informatics; 5) medical education; 6) public health; and 7) medical robotics. We examine their challenges, followed by a critical discussion about potential future directions and pitfalls of large AI models in transforming the field of health informatics.

Maciej A. Mazurowski, Haoyu Dong, Hanxue Gu et al.

Training segmentation models for medical images continues to be challenging due to the limited availability of data annotations. Segment Anything Model (SAM) is a foundation model that is intended to segment user-defined objects of interest in an interactive manner. While the performance on natural images is impressive, medical image domains pose their own set of challenges. Here, we perform an extensive evaluation of SAM's ability to segment medical images on a collection of 19 medical imaging datasets from various modalities and anatomies. We report the following findings: (1) SAM's performance based on single prompts highly varies depending on the dataset and the task, from IoU=0.1135 for spine MRI to IoU=0.8650 for hip X-ray. (2) Segmentation performance appears to be better for well-circumscribed objects with prompts with less ambiguity and poorer in various other scenarios such as the segmentation of brain tumors. (3) SAM performs notably better with box prompts than with point prompts. (4) SAM outperforms similar methods RITM, SimpleClick, and FocalClick in almost all single-point prompt settings. (5) When multiple-point prompts are provided iteratively, SAM's performance generally improves only slightly while other methods' performance improves to the level that surpasses SAM's point-based performance. We also provide several illustrations for SAM's performance on all tested datasets, iterative segmentation, and SAM's behavior given prompt ambiguity. We conclude that SAM shows impressive zero-shot segmentation performance for certain medical imaging datasets, but moderate to poor performance for others. SAM has the potential to make a significant impact in automated medical image segmentation in medical imaging, but appropriate care needs to be applied when using it.

Yongsong Huang, Wanqing Xie, Mingzhen Li et al.

Federated learning (FL) enables multiple client medical institutes collaboratively train a deep learning (DL) model with privacy protection. However, the performance of FL can be constrained by the limited availability of labeled data in small institutes and the heterogeneous (i.e., non-i.i.d.) data distribution across institutes. Though data augmentation has been a proven technique to boost the generalization capabilities of conventional centralized DL as a "free lunch", its application in FL is largely underexplored. Notably, constrained by costly labeling, 3D medical segmentation generally relies on data augmentation. In this work, we aim to develop a vicinal feature-level data augmentation (VFDA) scheme to efficiently alleviate the local feature shift and facilitate collaborative training for privacy-aware FL segmentation. We take both the inner- and inter-institute divergence into consideration, without the need for cross-institute transfer of raw data or their mixup. Specifically, we exploit the batch-wise feature statistics (e.g., mean and standard deviation) in each institute to abstractly represent the discrepancy of data, and model each feature statistic probabilistically via a Gaussian prototype, with the mean corresponding to the original statistic and the variance quantifying the augmentation scope. From the vicinal risk minimization perspective, novel feature statistics can be drawn from the Gaussian distribution to fulfill augmentation. The variance is explicitly derived by the data bias in each individual institute and the underlying feature statistics characterized by all participating institutes. The added-on VFDA consistently yielded marked improvements over six advanced FL methods on both 3D brain tumor and cardiac segmentation.

Luisa Gallee, Meinrad Beer, Michael Goetz

Interpretability is often an essential requirement in medical imaging. Advanced deep learning methods are required to address this need for explainability and high performance. In this work, we investigate whether additional information available during the training process can be used to create an understandable and powerful model. We propose an innovative solution called Proto-Caps that leverages the benefits of capsule networks, prototype learning and the use of privileged information. Evaluating the proposed solution on the LIDC-IDRI dataset shows that it combines increased interpretability with above state-of-the-art prediction performance. Compared to the explainable baseline model, our method achieves more than 6 % higher accuracy in predicting both malignancy (93.0 %) and mean characteristic features of lung nodules. Simultaneously, the model provides case-based reasoning with prototype representations that allow visual validation of radiologist-defined attributes.