Big Data, Small Training Sets: Addressing Medical Image Segmentation Bottlenecks

Recent progress in medical imaging technology leads to a continuous growth in clinical data volumes, which requires the development of Big Data tools to process and analyze the massive imaging data. Parallel processing and advanced machine learning techniques brought to bear on addressing fundamental medical image computing challenges such as segmentation. Nevertheless, the main bottleneck, the lack of annotated data to be used for the training, still remains. In my talk I will present unsupervised and semi-supervised techniques to accommodate medical image segmentation in the era of Big Data. Specifically, I will introduce a segmentation framework for large image ensembles accounting for multi-modal, multi-region and longitudinal MRI brain scans.