Advanced segmentation techniques for lung nodules, liver metastases, and enlarged lymph nodes in CT scans

This article presents advanced algorithms for segmenting lung nodules, liver metastases, and enlarged lymph nodes in CT scans. Segmentation and volumetry are essential tasks of a software assistant for oncological therapy monitoring. Our methods are based on a hybrid algorithm originally developed for lung nodules that combines a threshold-based approach with model-based morphological processing. We propose extensions that deal with particular challenges of each lesion type: lung nodules that are attached to non-convex parts of the pleura, rim-enhancing and peripheral liver metastases and lymph nodes with an extensive contact to structures of similar density. We evaluated our methods on several hundred lesions in clinical datasets and the quality of segmentations was rated by radiologists. The results were classified as acceptable or better in 81% to 92% of the cases for the different algorithms and readers.