AbstractAimto evaluate validity of digital quantification when compared with human fast-scoring in routine classification of chronic gastric wall inflammation.Method87 bariatric gastric samples coming from asymptomatic severe obese patients were examined and classified as normal, with unspecified chronic gastritis and lymphocytic gastritis using a fast-scoring, visual analogue scale method. Results were compared with digital diagnostic data (manual segmentation, supervised learning analysis based on intraepithelial lymphocytes count criteria). Discordant results were re-evaluated by the human pathologist by direct count (ground truth). Helicobacter Pylori diagnostic was performed in all cases (Giemsa).ResultsDigital analysis classified chronic inflammation as lymphocytic gastritis in 45 cases (mean 53 lymphocytes / 100 epithelial gastric cells ±18). 30 cases were labeled as unspecific chronic gastritis (mean 25/100±2.8) (p<0.0001). Human fast-scoring classified 43 cases as lymphocytic gastritis and 20 as unspecific gastritis. Helicobacter Pylori was detected in 49% of lymphocytic gastritis cases and in 7% of chronic gastritis. 47 diagnostics were concordant (54%). In 36%, digital score was better than human fast-scoring. In 7%, digital results were false negative (all cases generated by technical artifacts). Overall, digital quantification had 89% accuracy and 96% precision when compared with ground truth.ConclusionIn our study, digital image analysis produced a fast and reproducible classification of chronic gastric inflammation with good precision and accuracy. Technical errors generated 6 cases of false negative results. Several other limitations of the study (use of only bariatric gastric fundus tissue, low number of cases, manual supervised learning segmentation) ask for an increased number of cases evaluation before validation and clinical use.