Our paper was published in “Scientific Reports” by Nature Research


We are delighted to share that our manuscript entitled*“Deep learning models for histopathological classification of gastric and colonic epithelial tumours”*by Koji Arihiro M.D., Ph.D. and Kei Kato from Hiroshima University along with Medmain members Osamu Iizuka, Fahdi Kanavati, Michael Rambeau, Masayuki Tsuneki, was published in authoritative Scientific Reports, a journal of the*Nature Research*on Jan 30th.

Link:https://www.nature.com/articles/s41598-020-58467-9

DOI: 10.1038/s41598-020-58467-9

Pubmed link:https://www.ncbi.nlm.nih.gov/pubmed/32001752

The number of world cancer cases has been increasing. According to statistics, stomach and colon cancers are amongst the most common causes of cancer deaths in the world. Therefore, the cases of histopathological classification of those fields, performed by pathologists is in high demand.

Globally, there is a chronic shortage of pathologists and is getting more serious. This is one of the reasons that expectations on computational pathology techniques based on AI has been in more demand than ever before.

▶AI Pathological Classification with High Accuracy

In this study, our team has trained convolutional neural networks (CNNs) and recurrent neural networks (RNNs) on biopsy histopathology whole-slide images of stomach and colon. The models were trained to classify whole-slide images into various types of tumors such as adenocarcinoma, adenoma, and non-neoplastic. We evaluated our models on three independent test sets each, achieving area under the curves (AUCs*) up to 0.97 and 0.99 for gastric adenocarcinoma and adenoma, respectively, and 0.96 and 0.99 for colonic adenocarcinoma and adenoma respectively.

The study concluded that the classification accuracy done by the AI model is high enough to be deployed in actual medical sites and to be utilized as a screening tool for histology judgment, as well as a double-check tool.

(*AUC ranges in value from 0 to 1. A model whose predictions are 100% wrong has an AUC of 0.0; one whose predictions are 100% correct has an AUC of 1.0.)

Medmain will soon officially launch the pathological analysis system “PidPort” .”PidPort” is a platform service for pathological image diagnosis, which includes Deep Learning driven immediate pathological analysis.We hope this will assist and alleviate medical practitioners’ harsh daily workflow. By leveraging the power of technology, we aim to create a society where everyone can attain immediate high standard pathological diagnosis.