Comprehensive Study of Brain Age Prediction using Classical Machine Learning and Neural Networks

Abstract: The biological age of the human brain is an important biomarker in inspecting and maintaining the health of an individual. The brain age provides insights into an individual’s brain health due to genetics, environment, and lifestyle. The application of emerging state-of-the-art machine learning and deep learning technologies in the medical imaging fields may lead to faster and more precise solutions to complex, time-heavy problems. This thesis project, elucidates the foundational concepts of brain anatomy, the role of neuroimaging, and the application of both traditional machine learning algorithms and modern deep learning algorithms to predict brain age. It explores the predictive capabilities of algorithms ranging from classical machine learning algorithms like linear regression and decision trees to state-of-the-art deep learning technologies such as U-Net. The evaluation of the models on their predictive capabilities was done using standard evaluation metrics such as Mean Square Error and Root Mean Square Error, providing insights into the nuanced relationship between the features and brain age. The results of the models showed a variance in predicted age and the chronological age upto ±7.786and ±7.704 and ±8.315 years using supervised machine learning, fully connected neural networks and convolution neural network methods respectively. Ultimately,the study aims to provide insights into the amalgamation of domain knowledge, and modern technologies in the field of neuroimaging.