Radiomics and Deep Learning in Clinical Imaging: What Should We Do?

Excerpt

During the past several years, radiomics and deep learning (DL) became hot issues in medical imaging field, especially in cancer imaging. Radiomics is an emerging field of medical imaging that uses a series of qualitative and quantitative analyses of high-throughput image features to obtain diagnostic, predictive, or prognostic information from medical images. Recently, radiomics methods have been used to analyze various medical images including CT, MR, and PET to provide information regarding diagnosis, patients’ outcome, tumor phenotypes, and the gene-protein signatures in various diseases including cancer. Texture analysis is one of representative methods in radiomics. Machine learning (ML), a subset of artificial intelligence (AI), is a series of methods that automatically detect patterns in data, and utilize the detected patterns to predict future data or to make a decision making under uncertain conditions. DL is a kind of ML, which originated from artificial neural network in 1950. After resolving several critical limitations, deep learning has been applied in medical field since the 2000s. The most representative characteristic of ML and DL is that it is driven by data itself, and the decision process is finished with minimal interaction with a human. The ML and DL program can learn by analyzing training data, and make a prediction when new data is put in. DL is suitable to draw useful knowledge from medical big imaging data. This new AI technology in medical imaging has a potential to perform automatic lesion detection for differential diagnoses and, also, to provide other useful information including therapy response and prognostication. In these aspects, both radiomics and DL are closely related to each other in medical imaging field. For example, the radiomics data can be easily analyzed and clinically applied by the DL method, which facilitate precision medicine. Figure 1 shows the recent dramatic increased publications regarding radiomics and DL in the imaging fields.

…