Enhanced Brain MRI disease classification via wavelet decomposition-infused CNN architecture featuring residual blocks

Abstract

Precisely identifying diseases from brain Magnetic Resonance Imaging (MRI) plays a pivotal role in both diagnosing medical conditions and formulating effective treatment plans although it faces challenges owing to the complexity of the brain.  Recent studies highlight CNNs' efficacy in disease classification. We propose a CNN model integrating wavelet decomposition and Residual blocks to classify Alzheimer's disease, brain tumors, and normal conditions from MRI scans. Four stages of wavelet decomposition extract features, aiding Residual blocks in efficient deep network training. Evaluation on MRI datasets shows high accuracy, specificity, and sensitivity (0.945, 0.985, and 0.945 respectively), surpassing existing models. This model enhances diagnosis and treatment planning efficiency. Exploring various wavelet types, Daubechies-9 (DB9) wavelet proves superior, emphasizing wavelet selection importance. The model excels in binary, three-way, and four-way classifications, showcasing its adaptability and potential in brain MRI analysis