Objective To develop an artificial intelligence diagnostic model based on convolutional neural network (CNN), which can automatically identify the lesion area of colorectal cancer and predict the tumor staging of colorectal cancer under endoscopic ultrasound. This method can avoid staging errors caused by human factors and increase the objectivity and uniformity of endoscopic ultrasound in identifying and staging of colorectal cancer lesions.

Methods A total of 1436 endoscopic ultrasound images of patients with colorectal cancer who underwent endoscopic ultrasound at the Endoscopy Center of Foshan Second People's Hospital from February 2022 to October 2024 were collected in this study, and randomly divided into the training set and testing set. Using the CNN algorithm as the basic architecture, images in the training set were utilized to construct a prediction model for the lesion and non-lesion area of colorectal cancer under endoscopic ultrasound as well as its staging. Images in the testing set were used to validate the model and evaluate the diagnostic effectiveness of the prediction model.

Results A total of 1436 endoscopic ultrasound images were included in this study, including 1275 in the training set and 161 in the testing set. In the training set, 84 images were classified as T1 stage, 273 as T2 stage, 806 as T3 stage, and 112 as T4 stage, respectively. In the testing set, this model can effectively distinguish between the lesion and non-lesion area, with an accuracy of 82.0%. The accuracy of T2 and T3 stages in this prediction model was 82.9% and 88.9%, and 37.5% and 0 for T1 and T4 stages, respectively.

Conclusions In this study, a CNN-based model is developed for identifying the lesion and non-lesion area of colorectal cancer as well as T staging under endoscopic ultrasound. This model can effectively distinguish the lesion from non-lesion areas, which performs better in T2 and T3 stages compared with T1 and T4 stages. The overall predictive ability of this model for T stage has certain limitations, which requires further improvement and validation.