Skip to main content
ENDO 2026 Main banner

Neuroendocrinology and Pituitary

SAT-088 - Efficientnetb0 Powered Cross-Platform Application Enables Rapid and Accurate Detection and Differential Diagnosis of Pituitary Tumors from Brain Magnetic Resonance Imaging

Saturday, June 13, 2026
9:30 AM - 4:30 PM CT
Location: ENDOExpo; Poster Floor
Disclosure(s):
Gowrishankar Palaniswamy, MD: No financial relationships to disclose
Elangovan Krishnan, MBBS, M.Tech., MS, PhD: No financial relationships to disclose
Aboorva K Sudhakar, MD: No financial relationships to disclose
Introduction

Pituitary tumors are common neuroendocrine neoplasms with heterogeneous imaging appearances and variable clinical impact depending on hormonal activity and mass effect. Accurate MRI-based diagnosis is essential for timely endocrine and neurosurgical management but is subject to interobserver variability and resource constraints. Computationally efficient artificial intelligence models may improve diagnostic consistency while enabling scalable deployment in routine clinical settings.

Aim

To develop and externally validate a computationally efficient deep learning model for automated detection of pituitary tumors and differentiation from other intracranial neoplasms on brain MRI.

Methods

We developed a convolutional neural network based on EfficientNetB0, selected for its optimal accuracy-to-computation balance achieved through compound scaling of network depth, width, and resolution. Multicenter brain MRI datasets were curated and harmonized, including pituitary tumors, meningiomas, aggressive gliomas, and normal controls. Images were independently annotated by at least two expert neuroradiologists to establish ground truth. Data were partitioned into training, validation, and independent external test cohorts in a 3:1:1 ratio. Transfer learning with pretrained weights and end-to-end fine-tuning were applied. Model performance was assessed using accuracy, sensitivity, specificity, precision, F1-score, F2-score, and area under the receiver operating characteristic curve (AUROC). The trained model was deployed in a cross-platform application and externally evaluated for inference speed, clinical usability, and diagnostic concordance by 27 international physicians.

Results

The model demonstrated strong discrimination between pituitary tumors, normal anatomy, and other intracranial neoplasms. On independent testing, overall accuracy exceeded 97 percent, with sensitivity for pituitary tumor detection above 96 percent and specificity greater than 98 percent. AUROC values exceeded 0.98, with consistently high precision and F1- and F2-scores across tumor classes, including meningiomas and gliomas. Inference time was under one second per MRI study on standard hardware. Physician-based evaluation demonstrated high diagnostic concordance and seamless integration into clinical workflows.

Conclusion

EfficientNetB0 enables fast, accurate, and scalable MRI-based detection and differential diagnosis of pituitary tumors using limited computational resources. Robust external performance and clinician-validated deployment highlight its translational potential to support neuroendocrine imaging, reduce diagnostic variability, and facilitate timely clinical decision-making in pituitary disease.

*Unless otherwise noted, all abstracts presented at ENDO must not be released to the press or the public until the date and time of presentation. For oral presentations, the abstracts are embargoed until the session begins. The Endocrine Society reserves the right to lift the embargo on specific abstracts that are selected for promotion prior to or during ENDO.*