RESNET-53 for Extraction of Alzheimer’s Features Using Enhanced Learning Models

Abstract

Detecting Alzheimer's disease typically involves a combination of medical and cognitive assessments, neuro imaging, and sometimes genetic testing. Machine learning and artificial intelligence (AI) techniques are being applied to analyze neuro imaging data, genetic information, and clinical records to develop predictive models for Alzheimer's disease risk and early detection. Many AI models, particularly deep learning models, lack interpretability. Understanding how a model reaches a particular diagnosis or prediction can be challenging, which is a concern in the medical field where interpretability and transparency are crucial. CNNs typically learn features directly from data without prior feature engineering. While this is an advantage, it may also limit the exploration of specific features or biomarkers known to be associated with Alzheimer's disease. Medical images often require pre-processing steps, such as normalization, registration, and segmentation, before feeding them into CNNs. The effectiveness of CNNs may depend on the quality and accuracy of these pre-processing steps. The proposed methodology combines both CNN-based feature extraction and integrates adaptive filtering techniques to leverage the strengths of each method. This hybrid approach can lead to improved Alzheimer's disease detection by enhancing image quality and extracting relevant features for diagnosis. The combination of filtering techniques and CNNs allows the network to focus on relevant features while filtering out noise and irrelevant information. The proposed methodology integrates Gaussian filter with bilateral filter to produce an adaptive filter. Bilateral filtering adapts to the local image structure and content. By using it in combination with Gaussian filtering, the model can adaptively filter different regions of the image, optimizing the smoothing and enhancement process based on local features. This can lead to more effective and discriminative feature learning. Using the traditional CNN approaches the feature extraction has got nearly 57.78% accuracy but with the proposed model the accuracy has improved to 94.24%.