Deep Learning Radiomics Approach for Therapeutic Response Assessment in Colorectal Liver Metastases.

Patients suffering from colorectal cancer present with distant metastases at the initial diagnosis. Over 50% of patients with colorectal cancer develop liver metastases and 80% of them are not a candidate for hepatic resection which is considered the best curative treatment. Thus, chemotherapy is the available option for them before and after resection. However, the response rate to chemotherapy among patients varies and it ranges from 40-60% which may expose patients to toxic effects rather than benefiting from the treatment. Therefore, treatment response assessment to chemotherapy has vital importance to patients suffering from colorectal liver metastases. Many studies had been proposed to evaluate the predictive power of biomarkers obtained from molecular for assessing the treatment response. However, due to the high cost of reproducing these biomarkers and their low predictive power in assessing the treatment response for different cancer types, none of these studies deployed in clinical practice yet. Recently, the radiomics approach has evolved to assess the therapeutic response by extracting quantitative features from routine radiological imaging such as CT, MRI, and PET scan. Studies have shown that radiomics is better than the reference standard (RECIST) that relies on assessing the response to chemotherapy by evaluating the tumor size rather than considering the spatial heterogeneity of metastatic lesions. Although radiomics plays a great role in disease classification, diagnosis, and treatment personalization in different cancer types, combining it with deep learning is promising as stated by many preliminary studies that focused more on their technical feasibility. So, there is a lack of studies that validate the predictive power of the deep learning radiomics approach for therapeutic response assessment in colorectal liver metastases.