Efficient Convolutional Neural Networks for Automated Cognitive Diagnosis

Digitization has transformed diagnostic methods in several healthcare sectors. The standard cognitive assessment tests, evaluate cognitive impairment including early stages that can potentially progress to Alzheimer's Disease. However, it poses challenges due to manual administration. Here we propose using a novel convolutional neural network described here as CogniNet and compare its performance with leading doodle recognition transfer learning models to automate the visuospatial aspect of cognitive tests. Based on our CogniNet model we developed a web-application on the Laravel framework with enhanced accessibility and security features. Our convolutional neural network achieved 91.5% accuracy, while the EfficientNet And MobileNet transfer learning models reached 87.5% and 85.5% respectively.