Artificial Intelligence in EUS for Diagnosing Pancreatic Solid Lesions

Utilization of Artificial Intelligence for the Development of an EUS-convolution Neural Network Model Trained to Differentiate Pancreatic Cancer From Other Pancreatic Solid Lesions

We aim to develop an EUS-AI model which can facilitate clinical diagnosis by analyzing EUS pictures and clinical parameters of patients.

EUS is considered to be a more sensitive modality than CT in detecting pancreatic solid lesions due to its high spatial resolution. However, the diagnostic performance is largely dependent on the experience and the technical abilities of the practitioners. Therefore, we aim to develop an objective EUS diagnostic model based on the convolutional neural network, an artificial intelligence technique. In addition, clinical parameters such as risk factors, tumor biomarkers and radiology findings are also added to this artificial intelligence model in order to mimic the actual clinical diagnosis procedures and to increase the performance of this model.

Inclusion Criteria:

• Patients who underwent EUS using a curved line array echoendoscope (GF-UCT260; Olympus Medical Systems) since 2014 in our affiliation.
• For each patient, all available native EUS pictures are included.
• Patients' diagnosis are validated by surgical outcomes or fine-needle aspiration (FNA) findings and have a compatible clinical course with a follow-up period of more than 6 months.

Exclusion Criteria:

• The image is of poor quality.
• The images contain unique marks which can potentially bias the model, such as the biopsy needle.