Silent Evaluation of DeepTRG for Assessing Tumor Response After Preoperative Treatment in Stomach and Gastroesophageal Junction Cancer

This study will evaluate whether an artificial intelligence (AI) system called DeepTRG can reliably assess cancer tissue removed during surgery after treatment given before surgery. It will include adults with stomach cancer or cancer at the junction between the stomach and the food pipe who have received treatment before surgery. Pathologists are doctors who examine tissue to diagnose disease. After cancer surgery, they assess how much living cancer remains and whether cancer is present in nearby lymph nodes. DeepTRG analyzes digital images of tissue slides to measure remaining cancer and describe tissue changes, such as scarring and inflammation. It combines information from the original tumor site and lymph nodes to produce a structured assessment. The main question is: \* How often can DeepTRG complete the required assessment and produce a clear result without a person correcting the AI analysis? The study will also examine: * How closely the system's measurements agree with assessments made by independent expert pathologists. * How accurately it detects remaining cancer, including small amounts of cancer and cancer in lymph nodes. * How long the analysis takes and how often it fails or produces an uncertain result. * Whether its performance differs across treatment types and cancer tissue types. This study will take place at the Fourth Hospital of Hebei Medical University in China. Eligible patients will be enrolled consecutively. Researchers will use tissue collected during the patients' planned surgery. No additional surgery or biopsy is required for this study. DeepTRG will run in the background in "silent mode." The doctors responsible for the patients' routine pathology reports and treatment decisions will not receive its results. Patients will continue to receive usual care. The AI model and its grading rules will be fixed before validation begins. Independent expert pathologists will assess the tissue without seeing the AI results. Researchers will then compare the expert assessments, routine pathology reports, and DeepTRG outputs. Cases with poor-quality images, failed analyses, or uncertain outputs will remain part of the assessment of how reliably the system works. Researchers will also collect information about subsequent treatment, cancer recurrence, and survival during follow-up. The initial analysis will focus on feasibility and measurement performance. A separate future study would be needed to determine whether using DeepTRG in clinical care improves doctors' decisions or patient outcomes.

Sex

All

Minimum age

18 Years

Inclusion Criteria:

* Age 18 years or older.
* Histologically confirmed adenocarcinoma of the stomach or gastroesophageal junction (Siewert type II or III), diagnosed before neoadjuvant treatment.
* Received neoadjuvant anticancer treatment before planned surgical resection.
* Scheduled to undergo surgical resection with curative intent at the Fourth Hospital of Hebei Medical University, with routine pathological examination of the primary tumor site and regional lymph nodes.
* Written informed consent obtained in accordance with the ethics-approved study protocol.

Exclusion Criteria:

* Non-adenocarcinoma malignancies, including squamous cell carcinoma, lymphoma, gastrointestinal stromal tumor, or neuroendocrine carcinoma.
* Surgery for recurrent gastric or gastroesophageal junction cancer.
* Previous inclusion of the patient's tissue or data in DeepTRG model training, grading-rule development, or parameter tuning.

Patients will not be excluded because of a complete pathological response, positive surgical margins, poor-quality or unavailable slides, digitization failure, algorithm failure, or uncertain or unassessable DeepTRG results. Technical limitations identified after enrollment will be recorded and retained in the feasibility analysis. Patients who do not proceed to surgical resection will be documented in the study flow and will not contribute to the specimen-based primary outcome denominator.