A Transfer Learning Approach to Classify the Brain Age from MRI Images

摘要

Predicting brain age from Magnetic Resonance Imaging (MRI) can be used to identify neurological disorders at an early stage. The brain contour is a biomarker for the onset of brain-related problems. Artificial Intelligence (AI) based Convolutional Neural Networks (CNN) is used to detect brain-related problems in MRI images. However, conventional CNN is a complex architecture and the time to process the image, large data requirement and overfitting are some of its challenges. This study proposes a transfer learning approach using Inception V3 to classify brain age from the MRI images in order to improve the brain age classification model. Models are trained on an augmented OASIS (Open Access Series of Imaging Studies) dataset which contains 411 raw and 411 masked MRI images of different people. The models are evaluated using testing accuracy, precision, recall, and F1-Scores. Results demonstrate that InceptionV3 has a testing accuracy of 85%. This result demonstrates the potential for InceptionV3 to be used by medical practitioners to detect brain age and the potential onset of neurological disorders from MRI images.