Detection of New and Expanding Lesions in Multiple Sclerosis Using Machine Learning

Multiple sclerosis (MS), a prevalent chronic inflammatory disease of the central nervous system, has seen a significant rise in cases over the past 15 years. Accurate diagnosis and monitoring, typically achieved through magnetic resonance imaging (MRI), are essential but labor-intensive. This study evaluates the efficacy of Jazz, a deep learning-based software, in automating the detection of new and expanding MS lesions in MRI scans. Using MRI data from 83 MS patients at two time points, we implemented two neural networks within Jazz: one optimized for high specificity and the other for high sensitivity. The high specificity network excelled at minimizing false positives, achieving a Dice coefficient above 0.742 for lesions over 100 voxels and missing no lesions larger than 150 voxels. The high sensitivity network aimed to reduce false negatives, showing a Dice coefficient peaking at 0.741 for lesions between 150-300 voxels, though it struggled with smaller lesions. The study focused on 20 patients who developed new lesions. Jazz significantly improved the efficiency and accuracy of MRI analysis by reducing cognitive load and potential errors. It facilitated seamless lesion annotation and report generation, demonstrating its potential to enhance clinical decision-making and patient outcomes. The high specificity network showed robust performance for larger lesions, while the high sensitivity network was effective in minimizing missed lesions, particularly for medium to large lesions. These findings highlight the clinical utility of AI-driven tools in MS management. By leveraging the strengths of both high specificity and high sensitivity networks, integrated with Jazz, the study demonstrated significant improvements in lesion detection accuracy and efficiency. The automated processes for image coregistration, contrast recognition, and lesion annotation substantially reduced the manual effort required, allowing radiologists to focus on interpretation and clinical decisions. Future research should focus on expanding the dataset to include more diverse patient populations and integrating multimodal imaging data to enhance lesion detection further. Prospective studies are also needed to assess the impact of AI-driven lesion detection on clinical decision-making, treatment outcomes, and patient quality of life. In conclusion, this study highlights the potential of AI-driven neural networks and advanced software tools in transforming MS management. By enhancing lesion detection accuracy and streamlining MRI analysis, these technologies can improve clinical outcomes and efficiency in healthcare settings. Further refinement and validation of these tools will help integrate them into routine clinical practice, leading to more personalized and precise care for MS patients.