Abstract
Disclosure: R. Elangovan: None. K. Elangovan: None. J.R. Sethuraj: None. E. Krishnan: None. G. Palaniswamy: None. S. Nanduri: None. K. Sakalabaktula: None. U. Qureshi: None. F. Rahman: None. Z. Baloch: None. Z. Ahmed: None. S. Biswas: None. R. Muralidhar: None. C. Robert: None. K.J. Patel: None. G. Rajamani: None. N. Vora: None. S. Malapati: None. K. Noor: None. M. Kaur: None. A.S. Agrawal: None. A. Franklin Johnson: None. P. Uvaraj: None. T. Dontulwar: None. F. Madhumithaa Jagannathan: None. P.C. Sirait: None. M. Amro Alrouh: None. S. Tamma: None. S.S. Bhurchandi: None. S.K. Bhurchandi: None. Introduction: Endocrine cancers present critical challenges due to their systemic hormonal effects and intricate diagnostic profiles. With nearly 10 million cancer-related deaths annually, innovative solutions are imperative. Artificial intelligence (AI) is revolutionizing endocrinology by achieving unparalleled diagnostic precision in endocrine malignancies, offering scalable tools to reduce diagnostic errors and improve outcomes. AIM: This study aimed to develop and validate cutting-edge AI models for endocrine tumor evaluation, using multimodal data, ensuring their global generalizability, and deployment feasibility through a privacy-preserving, universally accessible application to promote equitable cancer care. Methods: Advanced deep learning architectures (EfficientNets and ResNets) were rigorously tested on anonymized multimodal datasets (CT, MRI, ultrasonography, cytopathology, and histopathology) from endocrine tumors in the thyroid, pancreas, pituitary gland, adrenal glands, and ovaries. Internal evaluations ensured model reliability, while external testing across diverse datasets from six continents validated generalizability. An online platform was developed, enabling users to test the models with their own data locally. Real-world applicability was confirmed through independent assessments by healthcare professionals worldwide. Results:EfficientNets and ResNets achieved exceptional diagnostic performance across multiple endocrine tumors (AUROC >99%) and robust generalizability. EfficientNetB0 excelled in treatment response monitoring (>97%). When deployed in an online application, EfficientNetB0 exhibited superior computational efficiency, while maintaining rapid processing speeds (<1 second/image) and diagnostic accuracy. Multi-national healthcare professionals validated the online platform’s reliability and usability, underscoring its global applicability. Conclusion: This study demonstrates the transformative impact of AI in endocrine oncology. Our findings illustrate a paradigm shift in biomedical AI, where minimal computational resources yield maximal diagnostic performance. This highlights the potential of AI-driven approaches to optimize diagnostic workflows, even with minimal computational resources, revolutionizing biomedical AI applications in cancer care. APPLICATION: The AI-powered application provides a privacy-preserving, scalable solution for endocrine tumor detection, classification, treatment monitoring, and advanced clinical decision-making. Validated multi-nationally by healthcare professionals, it ensures robust real-world applicability, fostering equitable cancer care globally. Its adaptability across platforms positions it as a universal tool for cancer care worldwide. Presentation: Sunday, July 13, 2025