MO-A-207B-01: Radiomics: Segmentation & Feature Extraction Techniques.

Image segmentation is one of the core problems for applying radiomics-based analysis to images. However, achieving repeatable and accurate segmentations for large datasets is challenging. The choice of segmentation method, the metrics used to evaluate the quality of such segmentations all depend on the specific clinical problem. This course will introduce three approaches, namely, fully automatic, interactive, and semi-automatic methods for generating segmentations. The pros and cons of each approach and when to choose a specific method will be discussed. Evaluation and assessment of the quality of a segmentation method is essential before it can be deployed for high-throughput analysis such as radiomics. This course will present some of the metrics that can be used for assessing quality of segmentations and highlight their advantages and deficiencies. Another important issue with respect to generating high quality segmentations and ultimately extracting robust radiomics features is image pre-processing. A few pre-processing techniques that can be used to improve the robustness of the analysis for MR and CT images will be presented.

LEARNING OBJECTIVES: 1. Understand the difference and applicability of various segmentation methods. 2. Understand how pre-processing can be used to improve the robustness of feature extraction and segmentation 3. Understand some basics of evaluating the quality of segmentations and the relevance of such metrics for clinical problems.