Improving Stroke Detection and Patient Outcomes: A Modular Convolutional Neural Network Approach for Primitive Diagnosis of Cerebral Infarction in Non - Contrast Computed Tomography

The brain’s optimal functioning requires a continuous supply of oxygen-rich blood, delivered through arteries. Any restriction in this flow causes fast brain cell death owing to oxygen deprivation, potentially resulting in harm or demise in affected brain regions. This study introduce a novel Modular Convolutional Neural Network (MCNN) approach that combines CNN’s feature extraction capability with the K-Nearest Neighbor (K-NN) classification technique to detect cerebral infarction in non-contrast computed tomography (NCCT) brain images at an early stage. The use of K-NN enables efficient categorization of features extracted by the CNN, making the model robust in recognizing infarction patterns. This modular architecture provides flexibility in incorporating new learning strategies. Our classifier’s performance is measured using a variety of metrics. Despite severe data imbalance, the model achieved 92% accuracy with 94% positive predicted value (PPV), 93% sensitivity, and an F1-score. K-NN was chosen as the classifier since its performance coupled with cnn was greater than other ML classifiers.The model’s effectiveness suggests that this strategy may effectively detect early-stage infarctions, even tiny lesions that are difficult to recognize in NCCT images.