Volumetric Feature Learning for Query-by-Example in Medical Imaging Archives
Eduardo Pinho, João Silva, Carlos M. A. Costa · 2019
The increasing challenges and requirements of medical image retrieval systems are leading the scientific community towards exploring modern representation methods as a means to improve clinical information retrieval as we know it. While current research tackles medical image retrieval through text-based, visual-based, or mixed approaches, representation learning can play an important role in improving retrieval capabilities by encoding medical image content into compact representations, addressing the problem of dimensionality. This paper introduces the potential of representation learning for the retrieval of high dimensionality imaging studies through automatically learned representations for regions of interest. Preliminary results are presented for feature learning through adversarial auto-encoding, based on the VISCERAL medical image retrieval benchmark.