Neoadjuvant chemotherapy (NC) is an important treatment for advanced gastric cancer (AGC). However, tools that effectively predict the efficacy of NC before treatment are lacking. Computed tomography images before and after NC were used to construct a deep learning-based radiomics signature to predict the efficacy of NC, prognoses and postoperative adjuvant chemotherapy benefit.

Background: Neoadjuvant chemotherapy (NC) is an important treatment for advanced gastric cancer (AGC). However, tools that effectively predict the efficacy of NC before treatment are lacking.

Methods: Computed tomography images before and after NC were used to construct a deep learning-based radiomics signature to predict the efficacy of NC, prognoses and postoperative adjuvant chemotherapy benefit. Tumor regression grade (TRG) =0 or 1 was defined as a good response to neoadjuvant chemotherapy (GRNC), and TRG=2 or 3 was defined as a poor response to neoadjuvant chemotherapy (PRNC). 193 patients with AGC from January 2010 to December 2018 in two different China university hospitals were included in this study. The before neoadjuvant chemotherapy imaging scoring system (BNCISS), imaging change scoring system before and after neoadjuvant chemotherapy (ICSS), which were constructed based on computed tomography images before after treatment.

Inclusion Criteria:

• Prospective collection and a retrospective analysis of patients seen from January 2010 to December 2018 at Fujian Medical University Union Hospital (FMUUH) and the First Hospital of Lanzhou University (FHLU) who were diagnosed with gastric adenocarcinoma and underwent radical resection

Exclusion Criteria:

• Patients with concurrent malignant disease, distant metastases, palliative resection, or gastric stump cancer were excluded.

Prospective collection and a retrospective analysis of patients seen from January 2010 to December 2018 at Fujian Medical University Union Hospital (FMUUH) and the First Hospital of Lanzhou University (FHLU) who were diagnosed with gastric adenocarcinoma and underwent radical resection yielded 193 neoadjuvant chemotherapy (NC) patients, including 148 cases from FMUUH (derivation cohort) and 45 cases from the FHLU (validation cohort). Of the 148 patients from FMUUH, 139 had computed tomography (CT) images before NC, 141 had CT images after NC, and 132 had CT images before and after NC. All 45 patients in the FHLU underwent CT before and after NC. Patients with concurrent malignant disease, distant metastases, palliative resection, or gastric stump cancer were excluded.