Revolutionary artificial intelligence applications transform endocrine disease detection and management
Attendees at ENDO 2025 in San Francisco were all buzzing about the Elangovan family of Houston, Texas. That’s because high school-aged siblings, Ramya and Kavin, were presenting their research into how artificial intelligence (AI) could change the way certain endocrine disorders were diagnosed and treated. The Elangovans talk to Endocrine News about their research findings, the inspiration for this undertaking, as well as the technology’s potential for improving human health around the world.
The field of endocrinology stands at the precipice of a technological revolution. Breakthrough research presented at ENDO 2025 in San Francisco, Calif., demonstrates how artificial intelligence (AI) is democratizing expert-level diagnostics for diabetic retinopathy and endocrine cancers. Two pioneering studies, both led by the same team of researchers, showcase the extraordinary potential of AI-powered applications (apps) to address two pressing challenges in endocrine care. By leveraging deep learning algorithms that achieve near-perfect accuracy while operating on basic mobile devices, these apps show promise in making specialized care more accessible and bridging critical gaps in healthcare delivery, particularly in underserved regions worldwide.
A Family Affair
The research team that innovated these apps includes siblings Ramya Elangovan and Kavin Elangovan, two high-school scholars and the Lead Scientist in AI and Biomedical Engineering and the Lead Researcher in Technological Innovations at AIM Doctor in Houston, Texas, respectively, and their father and mentor Krishnan Elangovan, MB, BS, PGDHM, MTech, MS, PhD, executive director of AIM Doctor. “Carnegie Vanguard High School (CVHS) has been instrumental in enabling our research journey,” Ramya says. “It provided me with opportunities and academic facilities for constructing a robust intellectual foundation at the confluence of biology, advanced mathematics, and computer science.”
Their father, likewise, prefers to give credit where credit is due: “’Mentoring’ feels like too formal a word for the role I played. It was less didactic instruction and more principled stewardship. Ramya and Kavin, with incredible curiosity, charted their own technical course through the complex world of artificial intelligence. My role was to provide the moral and ethical compass for their journey.” The principles he sought to instill in his children were universal accessibility and an unwavering commitment to scientific rigor.
“The scale of Ramya and Kavin’s vision, to build universally accessible diagnostic tools for a myriad of health conditions far exceeded the resources or knowledge of any single family,” he explains. “AIM Doctor was our answer to that challenge. What began as a family dinner-table conversation, has since evolved into a global, collaborative ecosystem, a decentralized global platform dedicated to harnessing AI for the betterment of humanity.” AIM Doctor stands for “Artificial Intelligence in Medicine Doctor” (the family calls it “Ambitious Doctor” on a lighter note). “It is a non-profit initiative that was started by our family and has now grown into an international family of over 600 physicians, biomedical engineers, and AI scientists from six continents, all volunteering their time and intellect. Our mission is twofold: to create and to empower.”
“Growing up where the nearest doctor was a formidable journey away, I understood from a young age that a person’s health is inextricably linked to their access to care. This principle has been the bedrock of our family’s philosophy. When my children began their journey into AI, my role, as their mother and as a clinician, was to constantly anchor their brilliant technical explorations in the reality of patient need.” – Jansi Rani Sethuraj, BSN, RN, CCRN, University of Texas Health Sciences Center Houston, Texas
Attending ENDO 2025 was a huge first for the family. “It was a profoundly humbling and illuminating experience,” Ramya says. “To be in the presence of leading researchers and clinicians, the very people who are at the core of solving what the Endocrine Society calls ‘the most pressing health problems of our time,’ was extraordinary.” Her brother Kavin agrees, saying, “It reinforced the immense responsibility we have to ensure our work is not just technologically novel, but clinically relevant and genuinely useful. Being there, among thousands of experts dedicated to this mission, underscored that the challenges we face in medicine are so immense they transcend traditional boundaries of age and experience.”
The duo anonymized thousands of retinal and endocrine cancer images from diverse populations all over the world to train their apps. This approach addresses a critical limitation of many AI medical apps that lack diverse representation in their training data, a well-documented and significant issue in the field. Many AI medical apps have been trained primarily on data from certain demographic groups or geographic regions, which can lead to reduced accuracy and potential bias when applied to underrepresented populations. This has been particularly problematic in medical imaging AI, where factors like skin tone, genetic variations, and population-specific disease presentations can affect algorithm performance. Studies have shown that AI diagnostic tools trained on homogeneous datasets often perform poorly when applied to different ethnic groups or populations not well-represented in the training data. The Elangovans’ training approach, by contrast, ensures accurate performance across different ethnic groups and geographic regions.
“Listening to top endocrinologists from around the world discuss the clinical realities of managing diabetes, the complexities of endocrine cancer and other conditions, and the daily challenges patients face provided an invaluable, real-world context for the AI models we design,” Ramya says.
Retinal Screening App
Diabetic retinopathy remains one of the leading causes of preventable blindness globally, affecting more than 100 million people as diabetes prevalence continues to surge worldwide. The condition’s insidious progression often occurs without symptoms until irreversible vision loss has occurred, making regular screening crucial for preservation of sight. However, access to specialized ophthalmologic care remains limited in many regions, creating a critical gap between need and availability of expert assessment.
“A Simple Mobile Artificial Intelligence Retina Tracker (SMART) Powered by Efficient Deep Learning Models for Diagnosis and Prognosis of Diabetic Retinopathy” introduced the SMART app, which integrates deep-learning algorithms capable of analyzing retinal fundus images with extraordinary precision and may represent a paradigm shift in diabetic retinopathy screening.
SMART uses advanced architectures including EfficientNets, ResNets, and Vision Transformers. The EfficientNetB0 model demonstrated over 99% accuracy in detecting and staging diabetic retinopathy, processing each image in under one second. The system’s ability to differentiate diabetic retinopathy from 39 similar ocular conditions, including hypertensive retinopathy and retinal vein occlusion, adds important specificity to its diagnostic capabilities. This comprehensive differential diagnostic ability reduces false positives and enhances clinical confidence in the system’s recommendations. Perhaps most impressively, the system demonstrated superior computational efficiency, requiring only one-third the runtime of comparable ResNet18 architectures while maintaining robust generalization across all tested datasets with an area under the receiver operating characteristic curve (AUROC) exceeding 0.99.
“Ramya and Kavin, with incredible curiosity, charted their own technical course through the complex world of artificial intelligence. My role was to provide the moral and ethical compass for their journey. The scale of Ramya and Kavin’s vision, to build universally accessible diagnostic tools for a myriad of health conditions far exceeded the resources or knowledge of any single family.” – Krishnan Elangovan, MB, BS, PGDHM, MTech, MS, PhD, executive director, AIM Doctor, Houston, Texas
The app underwent rigorous validation using multiple established datasets including APTOS, JSIEC, IDRiD, and MESSIDOR, and healthcare professionals from multiple international institutions independently validated both its diagnostic reliability and user-friendly interface. Another aspect that sets SMART apart is its universal accessibility. The app operates effectively on basic smartphones and internet-connected devices, eliminating the need for specialized equipment or high-end computing resources. This democratization of expert-level diagnostic capability enables primary care providers to incorporate comprehensive eye examinations into routine diabetic care visits, while empowering ophthalmologists to streamline patient screening processes.
Endocrine Cancer Diagnosis App
Endocrine cancers affecting the thyroid, ovary, pancreas, pituitary, and adrenal glands present unique diagnostic challenges due to their complex hormonal effects and intricate clinical presentations. With approximately 10 million cancer-related deaths annually worldwide, the need for innovative, scalable diagnostic solutions is urgent. These malignancies often require specialized expertise for accurate diagnosis and staging, expertise that may not be readily available in many healthcare settings globally.
Like SMART, the app presented in “A Universally Accessible, Computationally Efficient, Artificial Intelligence Powered Application for Diagnosing Endocrine Cancers” leverages advanced deep-learning architectures, specifically EfficientNets and ResNets to analyze diverse medical imaging modalities including computerized tomography (CT) scans, magnetic resonance imaging (MRI), ultrasonography, and both cytopathology and histopathology images. The multimodal approach represents a significant advancement in AI medical apps, as it mirrors the comprehensive diagnostic process that endocrinologists and oncologists use in clinical practice.
Here again, the AI models demonstrated exceptional diagnostic accuracy, exceeding 99% in validation datasets across multiple endocrine cancer types, and healthcare professionals from multiple international institutions validated the platform’s reliability and usability. The EfficientNetB0 architecture proved particularly effective for treatment response monitoring, achieving over 97% accuracy in this critical clinical application. Like its retinal screening counterpart, the system processes each image in under one second, even on devices with limited computational resources.
The app’s ability to provide expert-level diagnostic support for multiple endocrine malignancies represents a significant advance in point-of-care oncology.
Inspirations
Ramya explains that her interest in medicine began far from AI: “As a child, my world revolved around the 64 squares of a chessboard.” She won the GM Susan Polgar Foundation 2018 National Chess Championships in the under-10 category and represented the United States at the 2018 World Cadet Chess Championships in Spain. Her dream at that time was to become a world champion, but a visit to her grandmother in India changed everything when she discovered that her grandmother had lost her vision to diabetic retinopathy. “I felt a profound sense of loss, and a new, deeper purpose began to take shape,” she says. “My grandma’s loss of vision was not just a personal loss; it opened my eyes to the inequities of healthcare across the world. She lost her sight not because her condition was untreatable, but because the diagnostic expertise she needed was inaccessible in the remote village where she lived. That realization shaped the very architecture of our work today. Even at that young age, it sparked in my mind that the solution couldn’t be another complex, expensive machine at a specialist’s office. It had to be universally accessible, computationally simple enough to operate in a basic smartphone, and intuitive enough for a primary care provider in a remote clinic. This crystallized into the defining mission of my life.”
AI came in not long after. Just as Ramya returned from India, the COVID-19 pandemic began. “In that uncertain time, I discovered the world of artificial intelligence,” she says. “Machine learning uses the same kind of pattern recognition, logical thinking, strategic discipline, and the ability to think through complex decision trees that I was trained in as a young child by chess. I quickly realized AI could help accomplish my life’s mission. But no one in my family was an expert in computer programing or AI, so I resolved to master these skills and bring my family along on the journey.”
And she didn’t stop there: “I curated a personal curriculum from some of the world’s premier open-courseware platforms, immersing myself in lectures from Harvard’s edX, MIT Open Courseware, and Stanford Professor Andrew Ng’s deep learning courses. But true understanding, for me, comes from practically doing and actively teaching. To test and deepen my own comprehension, I established a small AI co-laboratory in our home and began distilling complex concepts with my parents and younger brother. That act of teaching was the most rigorous test of my own knowledge. It transformed our home into a small, agile research incubator.”
Kavin, she explains, developed the critical insight to build the universally accessible web application that hosts their models. “This foundational learning then scaled to the project itself, which involved a three-year effort to source and curate anonymized retinal image datasets from six continents to ensure our models were free from the demographic biases that can limit the effectiveness of other AI tools,” she continues. “It was a progression from solitary learning to familial collaboration and finally to a global-scale biomedical engineering challenge.”
The siblings’ mother, Jansi Rani Sethuraj, BSN, RN, CCRN, of the University of Texas Health Sciences Center in Houston, presented their research at ENDO. “I am a nurse by profession, currently pursuing my Doctor of Nursing Practice (DNP) at the University of Texas Health Sciences Center at Houston, with a clinical focus on management and prevention of preventable diseases,” she says. “My DNP studies are deeply rooted in commitment to preventative care and health equity. This passion is not academic; it is personal. Growing up where the nearest doctor was a formidable journey away, I understood from a young age that a person’s health is inextricably linked to their access to care. This principle has been the bedrock of our family’s philosophy. When my children began their journey into AI, my role, as their mother and as a clinician, was to constantly anchor their brilliant technical explorations in the reality of patient need. I wanted to ensure that their work would serve the person who, like my own mother-in-law, faced the tragedy of preventable blindness simply because the right expertise was out of reach.”
Implications for Clinical Practice
These apps represent more than incremental improvements in medical AI. For one, they challenge the prevailing assumption that superior AI performance requires substantial computational resources, which opens new possibilities for resource-limited environments.
Secondly, their implications extend beyond individual patient care to population health management. The ability to screen people everywhere for diabetic retinopathy and endocrine cancers could significantly impact global disease burden, particularly in regions where early detection and intervention programs are currently lacking.
“It reinforced the immense responsibility we have to ensure our work is not just technologically novel, but clinically relevant and genuinely useful. Being there, among thousands of experts dedicated to this mission, underscored that the challenges we face in medicine are so immense they transcend traditional boundaries of age and experience.” – Kavin Elangovan, high-school scholar, Lead Researcher in Technological Innovations, AIM Doctor, Houston, Texas
Finally, both apps are designed with clinical workflow integration in mind. The SMART retinal tracker enables primary care providers to incorporate sophisticated eye examinations into routine diabetic care, potentially identifying sight-threatening complications before they become irreversible. The endocrine cancer diagnostic application provides rapid, expert-level analysis that can accelerate treatment decisions and reduce diagnostic errors. Both apps feature privacy-preserving architectures that enable local data processing, addressing concerns about patient data security and regulatory compliance across different healthcare systems. Moreover, the ultrafast processing speeds of both applications make them practical for busy clinical environments, while their high accuracy rates provide the reliability necessary for clinical decision-making. The ability to operate on standard mobile devices eliminates barriers to adoption and ensures that these tools can be implemented immediately in existing clinical workflows.
Ultimately, Ramya said it best: “The endocrine community’s willingness to engage with our research on diabetic retinopathy and endocrine cancer diagnostics felt like a powerful, collective call to action. It left us profoundly inspired to continue this work and contribute, however modestly, to the collective endeavor of advancing human health.”
A Synthesis of Perspectives
She furthermore leaves us with this to consider: “When I started research in AI in the field of medicine, I came to realize that the most pressing problems in medicine, particularly in diagnostics, often reside in the interstitial spaces between traditional disciplines. A pathologist understands tissue, a radiologist understands imaging, and a computer scientist understands algorithms, but the true breakthroughs seem to emerge from the synthesis of perspectives.”
To that end, Ramya consciously tailored her coursework to master these distinct yet deeply interconnected domains. “This interdisciplinary approach has been absolutely essential for my work,” she says. “It provided the biological context to understand the pathophysiology of retinopathy and the computational framework required to design and train state-of-the-art AI models. My aspiration has been to use my secondary education not merely as a time of learning, but as a deliberate preparation for functioning effectively in that challenging but incredibly fertile territory where medicine and AI converge. I believe this integrated foundation will be critical as I continue to pursue this mission in my future studies and beyond.”
“Listening to top endocrinologists from around the world discuss the clinical realities of managing diabetes, the complexities of endocrine cancer and other conditions, and the daily challenges patients face provided an invaluable, real-world context for the AI models we design.” – Ramya Elangovan, high-school scholar, Lead Scientist in AI and Biomedical Engineering, AIM Doctor, Houston, Texas
Aside from being the talk of the Moscone Center during ENDO 2025, Elangovan siblings’ innovative contributions have already received much acclaim. Most notably, their work was recently honored by Harvard Medical School and the New England Journal of Medicine AI, with a travel award to present at the 2025 SAIL (Symposium on Artificial Intelligence in Learning Health Systems) last May in Puerto Rico.
Horvath is a freelance writer based in Baltimore, Md. She wrote about the 2025 Transatlantic Alliance Award winner Ashley Grossman, FMedSci, in the July issue.