Recently, artificial intelligence (AI)-based technologies have been reported for the diagnosis of gastrointestinal diseases during upper and lower endoscopic procedures. The investigators developed and validated a new AI system with a deep neural network (DNN-CAD) to localize and identify early esophageal squamous cell cancer (ESCC) using white-light endoscopy.
This study was conducted in 2 centers in China utilizing 2428 (1332 abnormal and 1096 normal) white-light EGD images obtained from 746 patients. An 8-layer convolutional neural network was developed to classify normal and abnormal images. The accuracy of the DNN-CAD system was validated in a different dataset containing 187 images from 52 patients, which were reviewed by 16 endoscopists (senior, mid-level, and junior groups). In the validation dataset, DNN-CAD accurately identified 89 of the 91 early ESCC lesions, with a sensitivity, specificity, and accuracy of 97.8%, 85.4%, and 91.4%, respectively, and a positive predictive value and negative predictive value of 86.4% and 97.6%, respectively. The accuracy of the DNN-CAD system for early ESCC was similar to that of the senior endoscopists when compared to the mid-level and junior endoscopists (P<0.05). Junior endoscopists showed a significant improvement (20%) in the diagnostic sensitivity for early ESCC with the aid of DNN-CAD.
The results of this new DNN-CAD system show that machine-built algorithms can be used for the accurate diagnosis of early esophageal squamous cell cancer and can also help inexperienced endoscopists improve at detecting abnormal lesions using standard endoscopy. In the upcoming months/years, several more studies will be conducted validating and supporting the role of AI in GI endoscopy.
Cai SL, Li B, Tan WM, et al. Using deep learning system in endoscopy for screening of early esophageal squamous cell carcinoma (with video). Gastrointest Endosc
2019 July 11. (Epub ahead of print) (https://doi.org/10.1016/j.gie.2019.06.044