Suvrankar Datta

Suvrankar Datta

Auto-generated profile, based on publicly available information

Suvrankar Datta

Department of Radiodiagnosis and Interventional Radiology, All India Institute of Medical Sciences, New Delhi, India

Research Focus & Metrics

Academic Career Length

3Years (2024-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
4 July 2024 in Indian Journal of Radiology and Imaging
Verified
Journal Article
Open access
PDF file
PSPradosh Kumar Sarangi
Pradosh Kumar Sarangi
SDSuvrankar Datta
Suvrankar Datta
MSM. Sarthak Swarup
M. Sarthak Swarup
SPSwaha Panda
Swaha Panda
DNDEBASISH SWAPNESH KUMAR NAYAK
DEBASISH SWAPNESH KUMAR NAYAK
Himel MondalHimel Mondal
Himel Mondal
Background Artificial intelligence chatbots have demonstrated potential to enhance clinical decision-making and streamline health care workflows, potentially alleviating administrative burdens. However, the contribution of AI chatbots to radiologic decision-making for clinical scenarios remains insufficiently explored. This study evaluates the accuracy and reliability of four prominent Large Language Models (LLMs)—Microsoft Bing, Claude, ChatGPT 3.5, and Perplexity—in offering clinical decision support for initial imaging for suspected pulmonary embolism (PE). Methods Open-ended (OE) and select-all-that-apply (SATA) questions were crafted, covering four variants of case scenarios of PE in-line with the American College of Radiology Appropriateness Criteria. These questions were presented to the LLMs by three radiologists from diverse geographical regions and setups. The responses were evaluated based on established scoring criteria, with a maximum achievable score of 2 points for OE responses and 1 point for each correct answer in SATA questions. To enable comparative analysis, scores were normalized (score divided by the maximum achievable score). Result In OE questions, Perplexity achieved the highest accuracy (0.83), while Claude had the lowest (0.58), with Bing and ChatGPT each scoring 0.75. For SATA questions, Bing led with an accuracy of 0.96, Perplexity was the lowest at 0.56, and both Claude and ChatGPT scored 0.6. Overall, OE questions saw higher scores (0.73) compared to SATA (0.68). There is poor agreement among radiologists' scores for OE (Intraclass Correlation Coefficient [ICC] = −0.067, p = 0.54), while there is strong agreement for SATA (ICC = 0.875, p < 0.001). Conclusion The study revealed variations in accuracy across LLMs for both OE and SATA questions. Perplexity showed superior performance in OE questions, while Bing excelled in SATA questions. OE queries yielded better overall results. The current inconsistencies in LLM accuracy highlight the importance of further refinement before these tools can be reliably integrated into clinical practice, with a need for additional LLM fine-tuning and judicious selection by radiologists to achieve consistent and reliable support for decision-making.

26 March 2026 in NEJM AI
Verified
Journal Article
Closed access
HZHong-Yu Zhou
Hong-Yu Zhou
JAJulián Nicolás Acosta
Julián Nicolás Acosta
SASubathra Adithan
Subathra Adithan
SDSuvrankar Datta
Suvrankar Datta
ETEric J. Topol
Eric J. Topol
PRPranav Rajpurkar
Pranav Rajpurkar

21 August 2025 in Indian Journal of Radiology and Imaging
Verified
Journal Article
Open access
PDF file
SDSuvrankar Datta
Suvrankar Datta
PSPradosh Kumar Sarangi
Pradosh Kumar Sarangi
Dear Editor, The recent editorial by Alex and Kesavadas[1] eloquently highlights the transformative potential of large language models (LLMs) in radiological practice, from optimizing workflows to enhancing radiology education. Radiology departments globally are facing increasing demands, making the adoption of LLMs not just advantageous but inevitable. However, it is imperative that we also address critical considerations for the responsible deployment of these powerful tools into regular radiology workflows (see [Fig. 1] for an overview of major clinical LLM use cases in radiology and their evaluation metrics). A fundamental consideration when deploying LLMs in clinical practice involves choosing appropriate deployment strategies. While popular proprietary models such as GPT-4 demonstrate impressive clinical reasoning capabilities, reliance on external cloud-based services inherently requires the transfer of protected health information, often across national borders, posing significant privacy and regulatory challenges. India's recently enacted Digital Personal Data Protection (DPDP) Act, 2023, currently mandates explicit consent for data storage and limits unregulated cross-border data transfers. These provisions encourage strategies that retain radiology data within national boundaries, enhancing legal safety and operational simplicity. Conversely, smaller, domain-specific, locally hosted models, fine-tuned with de-identified data, present viable alternatives, offering comparable task-specific accuracy with reduced risks to patient data privacy and improved compliance with stringent data protection regulations which countries including India are increasingly adopting.[2] An equally important discussion is the necessity of seamless integration of LLM outputs with existing radiology information systems and electronic health records. The anticipated efficiency gains with LLM-driven tools may be undermined if radiologists need to navigate between disconnected interfaces. Thus, early and close collaboration between health care IT specialists, LLM developers, and radiologists is essential to ensure that LLM outputs are directly integrated into the existing clinical reporting interfaces.[3] Further complexity arises with the advent of agentic workflows, in which LLMs autonomously perform multi-step clinical tasks such as guideline retrieval, decision support, or automated follow-up scheduling. For instance, in a recent multi-reader pilot at our center, we deployed an agentic workflow for generating radiology reports, using four sequential LLM agents capable of decomposing dictated keywords, retrieving relevant radiographic features via web search, synthesizing key findings, and validating the final structured reports. The agentic pipeline significantly reduced the time to report generation, supporting the feasibility and efficiency of agentic reporting systems in real-world practice. Despite their potential to markedly reduce cognitive burden, these automated systems remain computationally costlier and propagate errors across multiple interdependent tasks, challenging conventional evaluation metrics.[4] Therefore, rigorous evaluation methods must move beyond traditional metrics like BLEU scores or token-matching accuracy. Comprehensive evaluation frameworks including specific ones for agentic workflows, such as those suggested by Jiang et al in their MedAgentBench paper[5] and combining with automated metrics, adversarial stress-testing also known as “red-teaming” in which multidisciplinary reviewers deliberately attempt to “jailbreak” an LLM to output unsafe recommendations, identify worst-case failure modes and identify biases, along with expert radiologist validation, are urgently required to robustly assess both clinical utility and safety prior to broad deployment. [Fig. 2] details a stage-wise evaluation framework for operational, research, and educational deployments of LLMs and agentic workflows in radiology. Continuous monitoring of the real-world performance of LLM agents, with iterative feedback loops, will remain critical to maintaining high clinical standards and reliability. Finally, while advocating for increased automation, it is paramount to maintain a symbiotic relationship between radiologists and LLM-driven software. Radiologists should always retain oversight of clinically impactful decisions, with automated outputs clearly being highlighted, and preferably indicating confidence intervals or uncertainty estimates, mandating radiologist review before clinical decisions are finalized. In conclusion, the radiology community must proactively and urgently adopt structured, phased strategies that address privacy concerns, ensure interoperability, rigorously validate emerging agentic tools, and support ongoing performance monitoring. These steps are essential to responsibly harness the full potential of LLMs in augmenting radiologic practice, while safeguarding patient privacy and well-being. Article published online: 21 August 2025 © 2025. Indian Radiological Association. This is an open access article published by Thieme under the terms of the Creative Commons Attribution-NonDerivative-NonCommercial License, permitting copying and reproduction so long as the original work is given appropriate credit. Contents may not be used for commercial purposes, or adapted, remixed, transformed or built upon. (https://creativecommons.org/licenses/by-nc-nd/4.0/) Thieme Medical and Scientific Publishers Pvt. Ltd. A-12, 2nd Floor, Sector 2, Noida-201301 UP, India

18 September 2025
Verified
Book Chapter
Closed access
PDF file
SPShripad Pate
Shripad Pate
AFAiman Farooq
Aiman Farooq
SDSuvrankar Datta
Suvrankar Datta
MSMusadiq Aadil Sheikh
Musadiq Aadil Sheikh
AKAtin Kumar
Atin Kumar
DMDeepak Mishra
Deepak Mishra

29 September 2026 in Indian Journal of Radiology and Imaging
Verified
Journal Article
Open access
SDSuvrankar Datta
Suvrankar Datta
PPPrerna Priyadarshini
Prerna Priyadarshini
ARAntarjot Kaur Rekhi
Antarjot Kaur Rekhi
Artificial intelligence (AI) is entering Indian radiology faster than the capacity of existing workflows to adapt, with increasing claims of expert-level algorithmic performance, a regulatory architecture beginning to form, and rising clinician interest. Yet translation of technically capable tools into reliable clinical value remains uneven. This perspective-style review synthesizes recent literature on radiology AI deployment, workflow integration, validation, and governance, with observations across public teaching hospitals, private diagnostic networks, and teleradiology settings. We argue that while the dominant discourse around medical AI remains algorithm-centric, the dominant problem in Indian radiology is workflow-centric. This mismatch is a major reason why promising tools often fail to deliver real-world value. We map the contemporary Indian radiology workflow, identify recurring structural bottlenecks, and locate the points at which AI can meaningfully augment clinical practice. We examine priority use cases according to clinical maturity, describe scenarios in which deployment may be unsafe or premature, and propose a staged departmental readiness framework aligned with India's emerging digital-health and AI-governance landscape. For radiologists, the article offers a practical way to evaluate AI tools against departmental realities. For administrators and policy stakeholders, it outlines a framework for institutional investment, validation, and governance. For developers and vendors, it clarifies where Indian clinical workflows may accept augmentation and where integration is likely to be difficult. Our central position is that Indian radiology departments should build workflow integration, local validation capacity, post-deployment audit mechanisms, and governance structures before acquiring AI tools, so that adoption is safe, measurable, and clinically meaningful.

16 February 2026 in Indian Journal of Radiology and Imaging
Verified
Journal Article
Open access
PDF file
PSPradosh Kumar Sarangi
Pradosh Kumar Sarangi
Himel MondalHimel Mondal
Himel Mondal
SDSuvrankar Datta
Suvrankar Datta
We appreciate Daungsupawong and Wiwanitkit's interest in our study titled “Evaluating ChatGPT-4's Performance in Identifying Radiological Anatomy in FRCR Part 1 Examination Questions.”[1] We would like to address the concerns raised regarding methodology and interpretation. The authors acknowledge the use of ChatGPT-4o (OpenAI) for correcting language and grammar. Article published online: 16 February 2026 © 2026. Indian Radiological Association. This is an open access article published by Thieme under the terms of the Creative Commons Attribution-NonDerivative-NonCommercial License, permitting copying and reproduction so long as the original work is given appropriate credit. Contents may not be used for commercial purposes, or adapted, remixed, transformed or built upon. (https://creativecommons.org/licenses/by-nc-nd/4.0/) Thieme Medical and Scientific Publishers Pvt. Ltd. A-12, 2nd Floor, Sector 2, Noida-201301 UP, India