Discovery Radiomics via Evolutionary Deep Radiomic Sequencer Discovery for Pathologically-Proven Lung Cancer Detection

摘要

While lung cancer is the second most diagnosed form of cancer in men andwomen, a sufficiently early diagnosis can be pivotal in patient survival rates.Imaging-based, or radiomics-driven, detection methods have been developed toaid diagnosticians, but largely rely on hand-crafted features which may notfully encapsulate the differences between cancerous and healthy tissue.Recently, the concept of discovery radiomics was introduced, where customabstract features are discovered from readily available imaging data. Wepropose a novel evolutionary deep radiomic sequencer discovery approach basedon evolutionary deep intelligence. Motivated by patient privacy concerns andthe idea of operational artificial intelligence, the evolutionary deep radiomicsequencer discovery approach organically evolves increasingly more efficientdeep radiomic sequencers that produce significantly more compact yet similarlydescriptive radiomic sequences over multiple generations. As a result, thisframework improves operational efficiency and enables diagnosis to be runlocally at the radiologist's computer while maintaining detection accuracy. Weevaluated the evolved deep radiomic sequencer (EDRS) discovered via theproposed evolutionary deep radiomic sequencer discovery framework againststate-of-the-art radiomics-driven and discovery radiomics methods usingclinical lung CT data with pathologically-proven diagnostic data from theLIDC-IDRI dataset. The evolved deep radiomic sequencer shows improvedsensitivity (93.42%), specificity (82.39%), and diagnostic accuracy (88.78%)relative to previous radiomics approaches.