Small Data Deep Learning for Lung Cancer Detection in CT

Big data analytics and deep learning (DL) for big data are ones of the most promising areas of research in healthcare and medicine. These technologies analyze large amount of complex heterogeneous data such as genomics, diagnostic, therapeutic, and electronic health records data; they are successful in improving healthcare of people, diagnostic and therapeutic performance of medical doctors, as well as reducing healthcare cost. In the area of DL for medical image diagnosis, however, collecting and annotating a large number of patient cases is a big challenge. DL models require medical images of10,000 to 100,000 patient cases to adequately train, which would take years to collect. In the past research, less attentions were given to DL models that did not require big data. In this research, we developed a DL model that can be trained with a small number of cases, which we call small-data DL. We investigated a required number of cases for the small-data DL model for computer-aided detection (CAD) of lung nodules in CT. CAD provides a decision-support service to radiologists. We demonstrated that our small-data DL model, massive-training artificial neural network (MTANN), was able to achieve a state-of-the-art performance with a small number of cases (>100). DL models that can be trained with a small number of cases would fill in the gap between big-data DL and areas where a large number of cases are not available in medicine and healthcare.