Post written by Zhen Li, MD, from the Department of Gastroenterology, Qilu Hospital of Shandong University, Jinan, China.

The focus of our study was to develop a convolutional neural network—based artificial intelligence (AI) system named iEUS to assist in diagnosing pancreatic neuroendocrine neoplasms (pNENs) and multiple types of pancreatic lesions (including pancreatic adenocarcinoma, autoimmune pancreatitis, and pancreatic cystic neoplasms) using EUS.

Endoscopic diagnosis of pNENs is operator-dependent and time-consuming, as pNENs can mimic the normal pancreas or other pancreatic lesions. Missed diagnosis or inappropriate therapy may lead to a prolonged disease course and increased patient costs. Consequently, there is a need to construct a “third eye” to assist endosonographers in achieving more accurate and efficient pNEN diagnosis.

Our results demonstrate the robust efficacy of iEUS in discriminating diverse pancreatic lesions. In a human—iEUS contest using prospectively collected EUS videos, iEUS exhibited excellent diagnostic performance for pNENs and multiple types of lesions, surpassing or matching the diagnostic accuracy of participating EUS novices, intermediate endosonographers, and EUS experts. Furthermore, with the assistance of iEUS, endosonographers across all experience levels showed improved diagnostic performance in pancreatic lesions.

In summary, iEUS holds significant potential in improving pNEN diagnosis of endosonographers in clinical practice, possibly ultimately improving outcomes of pNEN patients. In subsequent phases, we plan to incorporate additional EUS data encompassing various pancreatic diseases, along with multimodal information of patients, to further optimize iEUS functionality and accuracy.

Given that patient information, laboratory tests, imaging findings, and other EUS modalities (eg, color Doppler, elastography, and contrast-enhanced US) are all critical to endoscopic diagnosis in clinical practice, integrating such multimodal data could enhance the model’s diagnostic accuracy while mitigating the “black-box” limitation of AI through improved interpretability. Moreover, multicenter bedside clinical trials will be conducted to validate its performance in real-time use.

With the development of novel AI techniques, we anticipate that AI can be applied throughout the entire EUS examination, encompassing but not limited to endoscopic diagnosis, quality control, training, EUS-guided tissue acquisition, interventional therapies, automated reporting, and rapid on-site evaluation. The use of AI represents a milestone in advancing EUS applications. We remain committed to promoting the field of EUS through our contributions.

Graphical abstract

Read the full article online.

The information presented in Endoscopedia reflects the opinions of the authors and does not represent the position of the American Society for Gastrointestinal Endoscopy (ASGE). ASGE expressly disclaims any warranties or guarantees, expressed or implied, and is not liable for damages of any kind in connection with the material, information, or procedures set forth.