Monday, April 29, 2019 | 18:00 CET
Radiomics in lung cancer
Constance DeMargerie, Boston/US

Description
The term “radiomics” appeared for the first time in the scientific literature in 2012 and its occurrences have seen an extensive growth since. Radiomics consists in the computerized extraction of a large number of quantitative features from radiological images, in order to build diagnostic and predictive classifiers. Machine learning techniques are usually involved for data analysis. A radiomics study requires the following steps: image acquisition and preparation, lesion segmentation, feature extraction, feature selection, model building and model validation. Applications of radiomics in the thoracic oncology field are wide, and include the identification of malignant nodules in lung cancer screening programs, the assessment of histological subtypes in lung cancer, the diagnosis of tumor mutations (radiogenomics), and the prediction of tumor prognosis or of tumor therapeutic response. Thus, radiomics contributes to pave the way to less invasive diagnostic procedures and personalized therapies. However, some challenges remain. Radiomics studies require large datasets of clinical and radiological information. Moreover, variation of technical parameters during the image acquisition, lesion segmentation, and feature extraction steps limit the reproducibility of previously developed models. Deep learning methods, which extract directly the useful data from selected images without segmentation, will certainly enhance the future role of medical imaging in precision medicine.

Learning Objectives
• To understand the concept of radiomics and its technical foundations
• To discover the different applications of radiomics in the field of thoracic oncology
• To understand the main limitations of radiomics and the challenges to be addressed