A Novel Approach to Precision Diagnosis of Multiple Sclerosis Brain Lesions Utilizing a Convolutional-based Ensemble Classification Approach for Embedded Systems

• Published in: Proceedings (IEEE International Symposium on Embedded Multicore/Manycore SoCs. Online) pp. 447 - 454
• Main Authors: Chilaka, Saiakhil, Kapavarapu, Raghavendra Satwik, Rong, Zhikang
• Format: Conference Proceeding
• Language: English
• Published: IEEE 16.12.2024
• Subjects: Accuracy; axial scans; biomedical engineering; central nervous system; chronic autoimmune condition; clinical application; cognitive deficits; computational efficiency; Computational modeling; convolutional layers; convolutional neural network (cnn); data splitting; dataset; deep learning; diagnostics; early diagnosis; ensemble model; environmental factors; genetic factors; grayscale imaging; healthcare ai; healthcare technology; image preprocessing; image resizing; immunological factors; impaired vision; Lesions; limited mobility; machine learning; Magnetic resonance; Magnetic resonance imaging; medical imaging; model optimization; mri scans; multiclass stratification; Multicore processing; Multiple sclerosis; neurodegenerative disorder; neurological symptoms; performance metrics; personalized medicine; Planning; precision diagnosis; predictive modeling; receiving operating characteristic (roc); rgb conversion; sagittal scans; Testing; testing set; Training; training set; transfer learning; transfer learning application; treatment planning; validation set; vgg19
• ISSN: 2771-3075
• DOI: 10.1109/MCSoC64144.2024.00079

Multiple Sclerosis Is a chronic autoimmune condition characterized by an attack elicited through an immune response in the central nervous system. Affecting over 1.8 million people worldwide. The condition leads to a range of neurological symptoms, including limited mobility, impaired vision, and cognitive deficits. Although the exact causes of multiple sclerosis remain unclear, a combination of genetic, environmental, and immunological factors is believed to play an important role over multiple years. In such cases, an early and accurate diagnosis through MRI scans remains crucial for timely intervention and proper treatment. Leveraging a meticulously curated ensemble model, this approach integrates optimized convolutional layers with customized transfer-learning models, combining the strengths of both frameworks for enhanced performance. The obtained dataset contains sagittal and axial magnetic resonance scans (n=3427) in greyscale, obtained through a prior study that studied 131 subjects in 2021. In the pre-processing stage, all images were resized to 128 by 128 pixels for computational efficiency and converted to RGB, followed by two 80-20 splits, resulting in training, validation, and final testing sets. A final ensemble model consisting of an optimized VGG19 achieved an average 97.5% accuracy, and a receiving operating characteristic (ROC) of 97.8%. Subsequent analyses were directed towards intricate multiclass stratifications, elucidating avenues for nuanced diagnostics and treatment planning for multiple sclerosis.