Predictive Model for Assessment of Pathological Response of Colorectal Liver Metastases to Chemotherapy from CT Images

In this work, we propose a predictive deep learning model for assessment of pathological response of colorectal liver metastases to chemotherapy from CT images. We conducted a retrospective analysis of a prospectively maintained database of patients who underwent partial hepatectomy or biopsy of colorectal liver metastases. We introduce a novel variant of the Inception module that includes instance normalization layers to accommodate for various contrast agent timing and baseline examinations. The clinical Rubbia-Brandt tumor regression grade (TRG) obtained from histopathology images of the resected lesions was used as ground truth. For the most common TRG dichotomization, our model achieves an AUC of 0.87 ${\pm}$ 0.03. The results show that the model is able to establish a link between CT images and the pathological assessment.