Abstract 2014: Radiopathomics strategy combining multiparametric MRI with whole-slide image for pretreatment prediction of tumor regression grade to neoadjuvant chemoradiotherapy in rectal cancer

Backgrounds: Tumor regression grade (TRG) reflects the chemoradiosensitivity of rectal cancer patients undergone neoadjuvant treatment, and relates to distinct probabilities of cancer recurrence and survival. However, in clinical practice, it is significantly challenging to rely solely on radiographic or clinical diagnostic information to obtain a patient's pathological response pre-treatment for treatment optimization. The aim was combining both the tumor information of macroscopic radiological and microscopic pathological images to develop and validated a more accurate signature for pretreatment prediction of TRG to neoadjuvant chemoradiotherapy (nCRT) in patients with locally advanced rectal cancer (LARC) on a multicenter dataset. Materials and Methods: We prospectively enrolled 981 patients (303 in the primary cohort, 678 in three external validation cohort) with clinicopathologically confirmed LARC treated with nCRT followed by surgery between Aug 2007 and Nov 2017, and collected their pretreatment multi-parametric MRI (mp-MRI), whole-slide image (WSI) of biopsy specimen, and pathologic response according to AJCC TRG system as well as clinical outcomes. Briefly, an artificial intelligence model integrating quantitative imaging features of mp-MRI and WSI was proposed to predict each nCRT treated patient into a particular 4-category TRG. The signature from the model (hereafter Rp-Grade) was further assessed in three independent validation cohort. Results: Rp-Grade yielded an overall ACC of 87.76% and revealed significant improvement than signature generating from model with mp-MRI or WSI features alone in all validation cohorts (P