Purpose: Early identification of demyelinating lesions is crucial for timely intervention in multiple sclerosis (MS). This study evaluates the performance of a deep learning–enhanced MRI pipeline designed to improve lesion conspicuity and reduce interpretation time.
Methods: A retrospective dataset of 620 MRI exams was processed using a convolutional neural network optimized for T2-weighted and FLAIR sequences. Radiologist performance, lesion detection rate, and reading time were compared between conventional and enhanced MRI datasets.
Results: Deep learning processing improved lesion detection by 18% and reduced average reading time by 22%. Inter-reader agreement increased significantly (κ = 0.87 vs 0.72). No false-positive inflation was observed.
Conclusion: AI-enhanced MRI demonstrates significant potential to assist radiologists in early MS detection, improving both accuracy and workflow efficiency.
Keywords: multiple sclerosis, MRI, deep learning, lesion detection, artificial intelligence