Saptarshi Purkayastha
Permanent URI for this collection
Risks and Opportunities of AI Recognition of Patient Race in Medical Imaging
Previous studies in medical imaging have shown disparate abilities of artificial intelligence (AI) to detect a person's race, yet there is no known correlation for race on medical imaging that would be obvious to human experts when interpreting the images. His recent work published in Lancet Digital Health demonstrates that deep learning models have extremely high accuracy at identifying self-reported race from medical images such as X-rays, MRIs and CTs. This ability raises serious concerns among some researchers. Such software might group patients, or influence their care, by factoring in race. These AI models work very well on poor quality, distorted and even images where many parts of the image were deliberately cut out. These types of categorizations could lead to inequality in providing health care and making recommendations, and human decision makers might not understand how and why AI models are making the recommendations. Engineers, clinical researchers and informaticians need to get together to identify how AI models are able to have these superhuman capabilities.
Professor Purkayastha's translation of research into potential ways to identify and mitigate risks of deploying AI models in clinical practice to avoid racial issues in healthcare treatment is another example of how IUPUI's faculty members are TRANSLATING their RESEARCH INTO PRACTICE.
Browse
Browsing Saptarshi Purkayastha by Title
Now showing 1 - 10 of 63
Results Per Page
Sort Options
Item AI recognition of patient race in medical imaging: a modelling study(Elsevier, 2022-06) Gichoya, Judy Wawira; Banerjee, Imon; Bhimireddy, Ananth Reddy; Burns, John L.; Celi, Leo Anthony; Chen, Li-Ching; Correa, Ramon; Dullerud, Natalie; Ghassemi, Marzyeh; Huang, Shih-Cheng; Kuo, Po-Chih; Lungren, Matthew P.; Palmer, Lyle J.; Price, Brandon J.; Purkayastha, Saptarshi; Pyrros, Ayis T.; Oakden-Rayner, Lauren; Okechukwu, Chima; Seyyed-Kalantari, Laleh; Trivedi, Hari; Wang, Ryan; Zaiman, Zachary; Zhang, Haoran; BioHealth Informatics, School of Informatics and ComputingBackground Previous studies in medical imaging have shown disparate abilities of artificial intelligence (AI) to detect a person's race, yet there is no known correlation for race on medical imaging that would be obvious to human experts when interpreting the images. We aimed to conduct a comprehensive evaluation of the ability of AI to recognise a patient's racial identity from medical images. Methods Using private (Emory CXR, Emory Chest CT, Emory Cervical Spine, and Emory Mammogram) and public (MIMIC-CXR, CheXpert, National Lung Cancer Screening Trial, RSNA Pulmonary Embolism CT, and Digital Hand Atlas) datasets, we evaluated, first, performance quantification of deep learning models in detecting race from medical images, including the ability of these models to generalise to external environments and across multiple imaging modalities. Second, we assessed possible confounding of anatomic and phenotypic population features by assessing the ability of these hypothesised confounders to detect race in isolation using regression models, and by re-evaluating the deep learning models by testing them on datasets stratified by these hypothesised confounding variables. Last, by exploring the effect of image corruptions on model performance, we investigated the underlying mechanism by which AI models can recognise race. Findings In our study, we show that standard AI deep learning models can be trained to predict race from medical images with high performance across multiple imaging modalities, which was sustained under external validation conditions (x-ray imaging [area under the receiver operating characteristics curve (AUC) range 0·91-0·99], CT chest imaging [0·87-0·96], and mammography [0·81]). We also showed that this detection is not due to proxies or imaging-related surrogate covariates for race (eg, performance of possible confounders: body-mass index [AUC 0·55], disease distribution [0·61], and breast density [0·61]). Finally, we provide evidence to show that the ability of AI deep learning models persisted over all anatomical regions and frequency spectrums of the images, suggesting the efforts to control this behaviour when it is undesirable will be challenging and demand further study. Interpretation The results from our study emphasise that the ability of AI deep learning models to predict self-reported race is itself not the issue of importance. However, our finding that AI can accurately predict self-reported race, even from corrupted, cropped, and noised medical images, often when clinical experts cannot, creates an enormous risk for all model deployments in medical imaging. Funding National Institute of Biomedical Imaging and Bioengineering, MIDRC grant of National Institutes of Health, US National Science Foundation, National Library of Medicine of the National Institutes of Health, and Taiwan Ministry of Science and Technology.Item Artificial Intelligence for Global Health: Learning From a Decade of Digital Transformation in Health Care(arXiv, 2020) Mathur, Varoon; Purkayastha, Saptarshi; Gichoya, Judy Wawira; BioHealth Informatics, School of Informatics and ComputingThe health needs of those living in resource-limited settings are a vastly overlooked and understudied area in the intersection of machine learning (ML) and health care. While the use of ML in health care is more recently popularized over the last few years from the advancement of deep learning, low-and-middle income countries (LMICs) have already been undergoing a digital transformation of their own in health care over the last decade, leapfrogging milestones due to the adoption of mobile health (mHealth). With the introduction of new technologies, it is common to start afresh with a top-down approach, and implement these technologies in isolation, leading to lack of use and a waste of resources. In this paper, we outline the necessary considerations both from the perspective of current gaps in research, as well as from the lived experiences of health care professionals in resource-limited settings. We also outline briefly several key components of successful implementation and deployment of technologies within health systems in LMICs, including technical and cultural considerations in the development process relevant to the building of machine learning solutions. We then draw on these experiences to address where key opportunities for impact exist in resource-limited settings, and where AI/ML can provide the most benefit.Item AuthN-AuthZ: Integrated, User-Friendly and Privacy-Preserving Authentication and Authorization(IEEE, 2020-10) Phillips, Tyler; Yu, Xiaoyuan; Haakenson, Brandon; Goyal, Shreya; Zou, Xukai; Purkayastha, Saptarshi; Wu, Huanmei; BioHealth Informatics, School of Informatics and ComputingIn this paper, we propose a novel, privacy-preserving, and integrated authentication and authorization scheme (dubbed as AuthN-AuthZ). The proposed scheme can address both the usability and privacy issues often posed by authentication through use of privacy-preserving Biometric-Capsule-based authentication. Each Biometric-Capsule encapsulates a user's biometric template as well as their role within a hierarchical Role-based Access Control model. As a result, AuthN-AuthZ provides novel efficiency by performing both authentication and authorization simultaneously in a single operation. To the best of our knowledge, our scheme's integrated AuthN-AuthZ operation is the first of its kind. The proposed scheme is flexible in design and allows for the secure use of robust deep learning techniques, such as the recently proposed and current state-of-the-art facial feature representation method, ArcFace. We conduct extensive experiments to demonstrate the robust performance of the proposed scheme and its AuthN-AuthZ operation.Item Big Data Analytics for developing countries – Using the Cloud for Operational BI in Health(Wiley, 2013) Braa, Jørn; Purkayastha, SaptarshiThe multi-layered view of digital divide suggests there is inequality of access to ICT, inequality of capability to exploit ICT and inequality of outcomes after exploiting ICT. This is evidently clear in the health systems of developing countries. In this paper, we look at cloud computing being able to provide computing as a utility service that might bridge this digital divide for Health Information Systems in developing countries. We highlight the role of Operational Business Intelligence (BI) tools to be able to make better decisions in health service provisioning. Through the case of DHIS2 software and its Analytics-as-a-Service (AaaS) model, we look at how tools can exploit Cloud computing capabilities to perform analytics on Big Data that is resulting from integration of health data from multiple sources. Beyond looking at purely warehousing techniques, we suggest understanding Big Data from Organizational Capabilities and expanding organizational capabilities by offloading computing as a utility to vendors through cloud computing.Item Blood Glucose Level Prediction as Time-Series Modeling using Sequence-to-Sequence Neural Networks(CEUR Workshop Proceedings, 2020-08) Bhimireddy, Ananth; Sinha, Priyanshu; Oluwalade, Bolu; Gichoya, Judy Wawira; Purkayastha, Saptarshi; BioHealth Informatics, School of Informatics and ComputingThe management of blood glucose levels is critical in the care of Type 1 diabetes subjects. In extremes, high or low levels of blood glucose are fatal. To avoid such adverse events, there is the development and adoption of wearable technologies that continuously monitor blood glucose and administer insulin. This technology allows subjects to easily track their blood glucose levels with early intervention without the need for hospital visits. The data collected from these sensors is an excellent candidate for the application of machine learning algorithms to learn patterns and predict future values of blood glucose levels. In this study, we developed artificial neural network algorithms based on the OhioT1DM training dataset that contains data on 12 subjects. The dataset contains features such as subject identifiers, continuous glucose monitoring data obtained in 5 minutes intervals, insulin infusion rate, etc. We developed individual models, including LSTM, BiLSTM, Convolutional LSTMs, TCN, and sequence-to-sequence models. We also developed transfer learning models based on the most important features of the data, as identified by a gradient boosting algorithm. These models were evaluated on the OhioT1DM test dataset that contains 6 unique subject’s data. The model with the lowest RMSE values for the 30- and 60-minutes was selected as the best performing model. Our result shows that sequence-to-sequence BiLSTM performed better than the other models. This work demonstrates the potential of artificial neural networks algorithms in the management of Type 1 diabetes.Item Comparative Performance Analysis of Different Fingerprint Biometric Scanners for Patient Matching(IOS Press, 2017) Kasiiti, Noah; Wawira, Judy; Purkayastha, Saptarshi; Were, Martin C.; BioHealth Informatics, School of Informatics and ComputingUnique patient identification within health services is an operational challenge in healthcare settings. Use of key identifiers, such as patient names, hospital identification numbers, national ID, and birth date are often inadequate for ensuring unique patient identification. In addition approximate string comparator algorithms, such as distance-based algorithms, have proven suboptimal for improving patient matching, especially in low-resource settings. Biometric approaches may improve unique patient identification. However, before implementing the technology in a given setting, such as health care, the right scanners should be rigorously tested to identify an optimal package for the implementation. This study aimed to investigate the effects of factors such as resolution, template size, and scan capture area on the matching performance of different fingerprint scanners for use within health care settings. Performance analysis of eight different scanners was tested using the demo application distributed as part of the Neurotech Verifinger SDK 6.0.Item Comparison of Open-Source Electronic Health Record Systems Based on Functional and User Performance Criteria(The Korean Society of Medical Informatics, 2019-04) Purkayastha, Saptarshi; Allam, Roshini; Maity, Pallavi; Gichoya, Judy W.; BioHealth Informatics, School of Informatics and ComputingObjectives: Open-source Electronic Health Record (EHR) systems have gained importance. The main aim of our research is to guide organizational choice by comparing the features, functionality, and user-facing system performance of the five most popular open-source EHR systems. Methods: We performed qualitative content analysis with a directed approach on recently published literature (2012-2017) to develop an integrated set of criteria to compare the EHR systems. The functional criteria are an integration of the literature, meaningful use criteria, and the Institute of Medicine's functional requirements of EHR, whereas the user-facing system performance is based on the time required to perform basic tasks within the EHR system. Results: Based on the Alexa web ranking and Google Trends, the five most popular EHR systems at the time of our study were OSHERA VistA, GNU Health, the Open Medical Record System (OpenMRS), Open Electronic Medical Record (OpenEMR), and OpenEHR. We also found the trends in popularity of the EHR systems and the locations where they were more popular than others. OpenEMR met all the 32 functional criteria, OSHERA VistA met 28, OpenMRS met 12 fully and 11 partially, OpenEHR-based EHR met 10 fully and 3 partially, and GNU Health met the least with only 10 criteria fully and 2 partially. Conclusions: Based on our functional criteria, OpenEMR is the most promising EHR system, closely followed by VistA. With regards to user-facing system performance, OpenMRS has superior performance in comparison to OpenEMR.Item Conversion of JPG Image into DICOM Image Format with One Click Tagging(Springer, 2017) Oladiran, Olakunle; Gichoya, Judy; Purkayastha, Saptarshi; BioHealth Informatics, School of Informatics and ComputingDICOM images are the centerpiece of radiological imaging. They contain a lot of metadata information about the patient, procedure, sequence of images, device and location. To modify, annotate or simply anonymize images for distribution, we often need to convert DICOM images to another format like jpeg since there are a number of image manipulation tools available for jpeg images compared to DICOM. As part of a research at our institution to customize radiology images to assess cognitive ability of multiple user groups, we created an open-source tool called Jpg2DicomTags, which is able to extract DICOM metadata tags, convert images to lossless jpg that can be manipulated and subsequently reconvert jpg images to DICOM by adding back the metadata tags. This tool provides a simple, easy to use user-interface for a tedious manual task that providers, researchers and patients might often need to do.Item Critical Components of Formative Assessment in Process-Oriented Guided Inquiry Learning for Online Labs(ACPI, 2019) Purkayastha, Saptarshi; Surapaneni, Asha K.; Maity, Pallavi; Rajapuri, Anushri S.; Gichoya, Judy W.; BioHealth Informatics, School of Informatics and ComputingIn the traditional lab setting, it is reasonably straightforward to monitor student learning and provide ongoing feedback. Such formative assessments can help students identify their strengths and weaknesses, and assist faculty to recognize where students are struggling and address problems immediately. But in an online virtual lab setting, formative assessment has challenges that go beyond space-time synchrony of online classroom. As we see increased enrollment in online courses, learning science needs to address the problem of formative assessment in online laboratory sessions. We developed a student team learning monitor (STLM module) in an electronic health record system to measure student engagement and actualize the social constructivist approach of Process Oriented Guided Inquiry Learning (POGIL). Using iterative Plan-Do-Study-Act cycles in two undergraduate courses over a period of two years, we identified critical components that are required for online implementation of POGIL. We reviewed published research on POGIL classroom implementations for the last ten years and identified some common elements that affect learning gains. We present the critical components that are necessary for implementing POGIL in online lab settings, and refer to this as Cyber POGIL. Incorporating these critical components are required to determine when, how and the circumstances under which Cyber POGIL may be successfully implemented. We recommend that more online tools be developed for POGIL classrooms, which evolve from just providing synchronous communication to improved task monitoring and assistive feedback.Item Current Clinical Applications of Artificial Intelligence in Radiology and Their Best Supporting Evidence(Elsevier, 2020-11) Tariq, Amara; Purkayastha, Saptarshi; Padmanaban, Geetha Priya; Krupinski, Elizabeth; Trivedi, Hari; Banerjee, Imon; Gichoya, Judy W.; BioHealth Informatics, School of Informatics and ComputingPurpose Despite tremendous gains from deep learning and the promise of artificial intelligence (AI) in medicine to improve diagnosis and save costs, there exists a large translational gap to implement and use AI products in real-world clinical situations. Adoption of standards such as Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis, Consolidated Standards of Reporting Trials, and the Checklist for Artificial Intelligence in Medical Imaging is increasing to improve the peer-review process and reporting of AI tools. However, no such standards exist for product-level review. Methods A review of clinical trials showed a paucity of evidence for radiology AI products; thus, the authors developed a 10-question assessment tool for reviewing AI products with an emphasis on their validation and result dissemination. The assessment tool was applied to commercial and open-source algorithms used for diagnosis to extract evidence on the clinical utility of the tools. Results There is limited technical information on methodologies for FDA-approved algorithms compared with open-source products, likely because of intellectual property concerns. Furthermore, FDA-approved products use much smaller data sets compared with open-source AI tools, because the terms of use of public data sets are limited to academic and noncommercial entities, which precludes their use in commercial products. Conclusions Overall, this study reveals a broad spectrum of maturity and clinical use of AI products, but a large gap exists in exploring actual performance of AI tools in clinical practice.