YoloV10 based Graph Convolution Network towards Disease Stage Classification of Alzheimer Disease using Structural Magnetic Resonance Imaging

Alzheimer disease is a complex neurological disease occurs in brain which causes memory impairment and cognitive challenges. According of wealth health organization (WHO), it is considered as chronic disease as slowly degrade the brain function and it predominately affects the memory and thinking skills of the patients. Further it leads to loss of brain tissues on formation of amyloid plague and aberrant protein deposit which is considered as abnormal protein. Thus it requires special attention to discover the Alzheimer disease in early stages to reduce the oxidative damages. Those information causes continuous decline of cognitive function over time. In particular, many researchers have initiated early diagnosis methods of the Alzheimer diseases using Structured Magnetic Resonance Imaging to observe the structures and its cognitive function in the hippocampus regions of the brain. However those approaches fail to discover amyloid plague deposition as it helps to determine the stage of the disease accurately. In this paper, a yolov10 model has been employed to graph convolution network to consider the deposition of the aberrant protein and amyloid plague in the hippocampus regions of the brain which categories the stage of the Alzheimer disease. Initially preprocessing step is employed to enhance the contrast of image and to eliminate the noise and non-brain tissues like skull on the MRI slices. Those preprocessed image is applied to the Yolov10 model for segmentation of the disease region with bounding box representation. Bounding box of the disease region indicates the aberrant protein and amyloid plague deposits. Those segmented region is projected to the graph convolution network. In this architecture, segmented region is transformed into graph structure with node representing pixels and edges mentioning associations. Transformed graph data is processed using adjacency matrix in the convolution layer of the network to extract the features and establish a non-linear graph embedding. Further grapy pooling layer of the model process the non-linear graph embedded features to produce the linear high level graph features. Finally fully connected layer utilizes the softmax function to classify the linear graph embedded feature into diseases classes of Alzheimer disease into stages as Mild Cognitive Impairment, Moderate Cognitive Impairment and Severe Cognitive Impairment. Experimental analysis and performance analysis of the graph convolution network architecture is performed using structured magnetic resonance images acquired from GRAD dataset in the python environment. Model produces the 99.1% accuracy in multiple disease stages and it proved to be outperforming compared to conventional approaches.