SAT-835 - Efficient and Scalable Pituitary Tumor Detection using Artificial Intelligence: Performance of Resnet18 on Brain MRI

Introduction

Pituitary tumors are a common cause of endocrine dysfunction and frequently require imaging-based diagnosis to guide medical and surgical management. Radiologic differentiation from meningiomas, aggressive gliomas, and normal anatomy can be challenging, particularly in resource-limited settings. Artificial intelligence models that balance diagnostic accuracy with computational efficiency may support scalable and accessible neuroendocrine imaging.

Aim

To evaluate the performance, generalizability, and efficiency of the ResNet18 architecture for pituitary tumor detection and differentiation from other intracranial neoplasms and normal brain MRI.

Methods

We adopted ResNet18, a residual convolutional neural network that applies identity based shortcut connections to stabilize gradient propagation while maintaining shallow depth. This design supports efficient feature learning with reduced parameter burden and rapid convergence. We curated brain MRI images representing pituitary tumors, meningiomas, gliomas, and non tumor controls from multi institutional sources.
Each image was independently reviewed by at least two expert neuroradiologists to establish ground-truth labels. Data were partitioned into training, validation, and independent test cohorts in the ratio 3:1:1. Transfer learning with pretrained initialization and full network fine tuning were performed. Model performance was assessed using accuracy, precision, recall, F scores, class wise AUROC, and loss behavior. External validation and deployment testing were conducted using a cross platform research application.

Results

ResNet18 demonstrated stable learning and robust generalization across heterogeneous datasets. On validation, the model achieved 97.9 percent accuracy with balanced precision, recall, and F1 score (≈0.98). Class-wise discrimination was strong for pituitary tumors, gliomas, meningiomas, and non-tumor controls, with consistently high AUROC values. On independent testing, overall accuracy was 94.3 percent with an F1 score of 0.94, confirming reproducible performance outside the training environment. Efficient inference enabled rapid analysis suitable for cross-platform clinical deployment.

Conclusion

ResNet18 demonstrates that compact residual architectures can deliver high diagnostic accuracy for pituitary tumor detection and differential diagnosis from brain MRI without excessive computational demand. Its efficiency, robustness, and external validation support its potential role as a scalable decision-support tool within neuroendocrine imaging workflows.

*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.*