||CNN Model Performance Analysis on MRI Images of an OASIS Dataset for Distinction Between Healthy and Alzheimer’s Patients
||Goo-Rak Kwon; Bijen Khagi
|| Medical MRI; CNN; AlexNet; GoogLeNet; ResNet50; CAD
||In this paper, we present the performance of a medical image classification model pretrained on natural images. In addition, another model is scratch trained from available medical magnetic resonance images in order to get a comparative analysis. We perform shallow tuning and fine-tuning of the pretrained model (AlexNet, GoogLeNet, and ResNet50) in a bunch of layers in order to find the impact of each section of layers in the classification result. We use 28 normal controls (NC) and 28 Alzheimer’s disease (AD) patients for classification, selecting 30 important slices from each patient. Once all the slices were collected, each model was trained, validated, and tested at a ratio of 6:2:2 on a random selection basis. The testing results are reported and analyzed so the final CNN model could be built with a minimal number of layers for optimal performance.