AI-based Diagnostic Tool Shows Promise in Upper GI Cancer

Article

A new study found artificial intelligence-based diagnostics could help with better diagnoses for upper gastrointestinal cancers.

A new system of artificial intelligence-based diagnostics using endoscopic imaging was highly accurate in diagnosing upper gastrointestinal cancers. The system, termed the Gastrointestinal Artificial Intelligence Diagnostic System (GRAIDS), could help community hospitals in improving gastrointestinal cancer diagnoses, according to the results of a study published in Lancet Oncology.          

“Most upper gastrointestinal cancers are diagnosed at advanced stages because their signs and symptoms tend to be latent and non-specific, leading to an overall poor prognosis, but if detected early, 5-year survival can exceed 90%,” wrote study authors led by Huiyan Luo, MD, of Sun Yat-sen University Cancer Center in Guangzhou, China. “Novel endoscopic strategies and techniques have been developed for these cancers, but they may not be effective in all areas. The risk of missing suspicious upper gastrointestinal cancers in endoscopy examinations might still be high in hospitals with low patient volume, in less developed or remote regions, and even in countries where many endoscopies are performed.”

 The researchers used a total of 1,036,496 endoscopy images from 84,424 individuals at 6 hospitals in China to develop and test the new GRAIDS method. They used only images from patients with histologically confirmed esophageal or gastric cancer and randomized individuals to training, intrinsic verification, and internal validation sets

 The GRAIDS algorithm was based on Google’s DeepLab V3+. Inputting the endoscopy images resulted in an output of whether or not the image contained a tumor, and then a segmentation of the tumor region of the image.

After development of the algorithm, GRAIDS achieved a diagnostic accuracy in identifying upper GI cancers in the internal validation set of 0.955. In 5 external validation sets, its accuracy ranged from 0.915 to 0.977. The algorithm’s diagnostic sensitivity was similar to that of an expert endoscopist, at 0.942 compared with 0.945 (P= 0.692). It was superior to that of a “competent” endoscopist (0.858; P< 0.0001) and a trainee (0.722; P< 0.0001).

The positive predictive value for GRAIDS was 0.814, compared with 0.932 for an expert endoscopist, 0.974 for a competent endoscopist, and 0.824 for a trainee. The negative predictive value for GRAIDS was 0.978, compared with 0.980, 0.951, 0.904 for the 3 endoscopists, respectively.

“To the best of our knowledge, this is the largest study in the field of artificial intelligence-guided cancer detection based on upper gastrointestinal endoscopic images worldwide,” the authors wrote. “GRAIDS could support non-expert endoscopists by improving their diagnostic accuracy to a level similar to that of experts.”

References:

Luo, H., Xu, G., Li, C., He, L., Luo, L., Wang, Z., Jing, B., Deng, Y., Jin, Y., Li, Y., Li, B., Tan, W., He, C., Seeruttun, S., Wu, Q., Huang, J., Huang, D., Chen, B., Lin, S., Chen, Q., Yuan, C., Chen, H., Pu, H., Zhou, F., He, Y. and Xu, R. (2019). Real-time artificial intelligence for detection of upper gastrointestinal cancer by endoscopy: a multicentre, case-control, diagnostic study. [online] Available at: https://www.thelancet.com/journals/lanonc/article/PIIS1470-2045(19)30637-0/fulltext [Accessed 9 Oct. 2019].

Recent Videos
According to Benjamin J. Golas, MD, PIPAC could be used as a bridging therapy before surgical debulking or between subsequent large surgical operations.
According to Benjamin Golas, MD, PIPAC is emerging as minimally invasive laparoscopic approach for patients with peritoneal carcinomatosis.
According to Ronan J. Kelly, deciding whether to give nivolumab- or durvalumab-based regimens in gastric cancers may rely on a patient’s frailty.
Five-year follow-up revealed that patients treated with nivolumab vs placebo in the phase 3 CheckMate 577 trial experienced a “doubling” of survival.
Patients treated with nivolumab in the phase 3 CheckMate 577 trial were less likely to experience progression-related treatment discontinuation vs placebo.
Testing a patient’s genetics may influence decisions such as using longer courses of radiotherapy, says Rachit Kumar, MD.
Spatial transcriptomics and multiplex immunohistochemistry from samples may elucidate outcomes for patients who undergo surgical care for cancer.
Future work may focus on optimizing symptom management associated with percutaneous transesophageal gastrostomy placement in malignant bowel obstructions.
Post-operative length of stay ranged from 4 to 9 days for patients who underwent percutaneous transesophageal gastrostomy for malignant bowel obstructions.
Related Content