Research of the Application of Pancreatic Cancer Screening Artificial Intelligence Model "PANDAPro"

Pancreatic Cancer

Artificial Intelligence, Pancreatic Cancer, PANDA PLUS, PANDA

The purpose of this study is to build upon the previously developed deep learning-based non-contrast CT pancreatic cancer screening model, PANDA. The model will first undergo training and enhancement, followed by external validation across multiple centers. Subsequently, a large-scale real-world validation will be conducted at Zhejiang University's First Affiliated Hospital , the study will be divided into two rounds. In the first round, the performance of the PANDA model and the upgraded PANDA Pro model will be compared on consecutive retrospective real-world CT scans. In the second round, physicians will record the PANDA Pro results in real time to identify potential pancreatic lesions that may have been clinically missed. By leveraging clinical big data across different scenarios at Zhejiang University's First Affiliated Hospital, the study aims to validate the model's role in prompting and supplementing the diagnosis of PDAC in clinical practice, thereby laying the foundation for large-scale opportunistic screening of PDAC.

In the first round of real-world validation, abdominal enhanced CT imaging data from patients treated at the First Affiliated Hospital of Zhejiang University between January 2018 and July 2022 were consecutively included. The results of both PANDA and PANDA Pro were recorded and compared with the pathological gold standard of the patients to evaluate model performance. In the second round of real-world validation, patients with non-contrast CT scans are expected to be enrolled. The enrolled patients will be categorized into three groups: nonPDAC, PDAC, and normal. The results will be compared with imaging report findings. Patients with PANDA Pro-reported PDAC positivity but no pancreatic lesions indicated in the imaging report, or those with positive pancreatic findings in the imaging report but no subsequent clinical intervention, will be identified as requiring follow-up. These patients will be recalled to the First Affiliated Hospital of Zhejiang University for further examination and diagnosis. Among them, patients with secondary examination confirming PDAC positivity will undergo standard clinical procedures such as MDT at the hospital, ultimately tracking their treatment to confirm ground truth. For patients with secondary examination indicating PDAC negativity, a preliminary determination of their true status will be made after a small-scale discussion between hepatobiliary surgeons and radiologists. Those judged as false positives will undergo follow-up for up to two years to determine their outcomes, thereby validating the sensitivity and specificity of the PANDA Pro model. For patients with PANDA Pro-reported nonPDAC positivity but no positive pancreatic findings in the imaging report, hepatobiliary-pancreatic surgeons and radiologists will retrospectively review the images to confirm their true status. For patients with PANDA Pro-reported normal results but imaging reports indicating pancreatic lesions, a secondary review by surgeons will be conducted to determine whether it is a false negative by PANDA Pro. For patients reported as true positives for PDAC by the PANDA Pro model, medical records (including tumor marker levels, patient symptoms, resectability classification, TNM staging, etc.) will be collected. These will be compared with corresponding indicators from PDAC patients identified through the SOC(Standard Order of Clinic) during the same period, to validate PANDA Pro's ability to detect and identify lesions in the early stages of pancreatic cancer development.

No related papers found yet.

China