Design, Development and Evaluation of a Deep Learning-Based Personalized Healthcare System for Diagnosis of Brain Metastases

Abstract

Brain metastases (BM), which affect 10-30% of cancer patients, represent important diagnostic and therapeutic problems due to their impact on cognitive function. Traditional manual MRI interpretation methods are time-consuming and potentially inaccurate, especially for tiny or diverse tumours. Artificial intelligence (AI) tools such as deep learning (DL) and machine learning (ML) made it possible to analyse complex MRI data quickly, accurately, and automatically, which was a major factor in BM diagnosis. This paper presents a novel approach for automatic brain metastases segmentation on MRI data that makes use of a U-Net model. To improve the accuracy of BM identification, the proposed method combines numerous imaging modalities, including T1-Weighted, T2-Weighted, T1-contrast enhanced, and Fluid-Attenuated Inversion Recovery (FLAIR). The University of California San Francisco Brain Metastases Stereotactic Radiosurgery (UCSF-BMSR) MRI dataset has been utilized for this purpose. The U-Net model was trained, verified, and tested on this dataset, and it performed admirably with an overall accuracy of 99.75%, a dice coefficient of 64.49%, and an Intersection over Union (IOU) of 96.81%. The proposed technique has been compared with two baseline models, namely Convolutional Neural Networks (CNN) and Fully Convolutional Networks (FCN). The U-Net model outperformed the baselines in all important measures, demonstrating its potential for real-world clinical application. The findings highlight the U-Net model’s capacity to greatly enhance BM detection accuracy, allowing for prompt treatment decisions.