Classification of Brain Infarction using Deep Learning techniques on Magnetic Resonance Imaging (MRI)

Background: This research concentrated on employing Convolutional Neural Networks (CNNs) to classify brain infarctions. A stroke occurs due to a blockage or hemorrhage in the blood vessels, which disrupts or diminishes the blood supply to the brain, leading to the death of brain cells. With the progress in machine learning techniques in medical imaging, early detection of stroke has become increasingly feasible, playing a crucial role in the diagnosis and treatment of this potentially fatal condition. Objective: Building an intelligent CNN model to classify types of brain infarction using up algorithms Median Filter preserving edges and efficient minimization of noise with durability against flowing noise, Fuzzy C – Mean to Image Segmentation and Recognition of Region of Interest for infarction location and size of the infarction for brain infarction Identification, GLCM, and FOSF. Methodology: The research was carried out in a private hospital; in Baghdad, Iraq, on 60 patients with brain infarction. Included were 9 hyperacute (>6 hours post brain infarction), 19 acute (7-72 hours), 12 subacute (4-7 days), and 20 chronic (< 15 days). We have introduced a Convolutional Neural Network (CNN) model as a solution to predict the likelihood of a patient experiencing a stroke at an early stage, aiming for maximize effectiveness and precision. Results: The ROC curve demonstrates that the average apparent diffusion coefficient (aveg (ADC) mm²/s) and relative apparent diffusion coefficient (rADC%) are two reliable and effective measures for monitoring brain infarction developments. High sensitivity indicates a strong ability to detect true positives (cases of infarction), while the shape and position of the curve suggest that average (ADC) mm²/s also maintains a reasonable specificity, minimizing false positives. This balance results in high overall diagnostic accuracy. Discussion: Both rACD (%) and aveg (ADC) mm²/s are robust indicators for the imaging monitoring of brain infarction developments. The inclusion of AI in the assessment process may result marginal improvements in specificity and accuracy.