Enhanced Image Retrieval and Classification Frameworks for Brain Disease Diagnosis Using Hybrid Deep Learning Models

Abstract

Brain disease diagnosis relies heavily on accurate image retrieval and classification, a task complicated by the vast volume and complexity of imaging data. This study presents a hybrid deep learning framework that combines convolutional neural networks (CNNs) for feature extraction with support vector machines (SVMs) for classification, optimized through transfer learning techniques. The proposed model was trained and tested on a dataset comprising MRI and CT scans of brain diseases, including Alzheimer’s and gliomas. Results demonstrated superior accuracy, precision, and recall compared to traditional and pure CNN models, with the hybrid model achieving over 90% accuracy in distinguishing disease-specific features. Visualization through annotated brain scans confirmed the model’s ability to identify critical disease markers, such as cortical thinning in Alzheimer's and abnormal tissue density in gliomas. Although limitations related to dataset diversity and image quality were observed, the model's performance suggests strong potential as a clinical diagnostic aid, capable of assisting radiologists with precise and timely brain disease classification. Future work will focus on expanding the model’s adaptability to a broader range of brain pathologies and diverse imaging conditions.