Researchers have developed a novel AI-powered online platform for diagnosing endocrine cancers with speed and accuracy. The AI models achieved diagnostic accuracy of 99% or more in recognizing multiple endocrine tumors. Reports of the development and validation of the models were presented at ENDO 2025, the annual meeting of the Endocrine Society (Abstract OR24-08).
“By democratizing access to advanced diagnostics, this AI innovation marks a paradigm shift in cancer care, offering hope for earlier detection, more precise treatment, and better survival for patients facing endocrine malignancies,” stated chief project mentor, Elangovan Krishnan, MBBS, PGDHM, MTech, MS, PhD, of AIM Doctor in Houston, Texas. “This AI-powered application can deliver fast, reliable, and affordable endocrine cancer diagnostics to anyone, anywhere, thereby helping to close gaps in cancer care and advance health equity worldwide.”
Study Methods and Rationale
The researchers sought out to develop an AI model for endocrine tumor evaluation in order to promote equitable cancer care. They hope that the resulting application can democratize specialized cancer diagnostics, making it more accessible for improved patient outcomes globally, even in resource-limited settings.
They tested two advanced deep learning architectures—EfficientNets and ResNets—on multimodal and anonymized datasets of computed tomography, magnetic resonance imaging, ultrasonography, cytopathology, and histopathology information from all forms of endocrine tumors to allow for maximum generalizability. Also, they curated images of endocrine cancers representing diverse global populations of patients with endocrine tumors to train and validate the models on these images for detection and staging purposes.
Then the model was developed into an online platform so that users could access and test the models with their local data. Health-care professionals across the world used and validated the model with their independent assessments to prove its real-world applicability.
Key Study Findings
The models achieved diagnostic accuracy above 99% (AUROC) with robust generalizability across multiple endocrine tumors.
As an online platform, the model demonstrated rapid processing speeds of less than a second per image while continuing to achieve diagnostic accuracy.
"Our findings illustrate a paradigm shift in biomedical AI, where minimal computational resources yield maximal diagnostic performance," the study authors wrote in their abstract.
Disclosure: For full disclosures of the study authors, visit endocrine.org.
