A transfer learning assisted 3D deep learning method for knee osteoarthritis detection

Abstract

The present invention discloses a transfer learning assisted 3D deep learning method (10) to detect and determine knee osteoarthritis (OA) from medical imaging. In a specific embodiment, the transfer learning assisted 3D deep learning method (10) comprising steps of: obtaining magnetic resonance imaging (MRI) volumes (11); converting the MRI volumes to Neuroimaging Informatics Technology Initiative (NIfTI) standard (12); transforming the MRI volumes to a standardized format that includes resizing, Z-normalizing, and standardizing (13); performing feature extraction from the MRI volumes using 13 different neural networks (14); training and comparing the MRI volumes using both pretrained and randomly assigned initial weights (15); using the pretrained weights as the initial weight of the training process of 3D models (16); and comparing and evaluating the performance of the 3D models based on classification performance metrics (17).