A Personalized Diagnostic Tool for Microbiome-Related Morbidities

Abstract

: A model-driven approach suitable for classifying microbiome-related morbidities such as ulcerative colitis on smart mobile devices is investigated in this manuscript. A novel scheme is proposed, which consists of a pre-trained image classifier on ImageNet and is deployed into the presented Android mobile application for this purpose. Endoscopic images of mouse colitis were used as input datasets for our experiments. The proposed approach offers an efficient classifier, based on the average of all its performance metrics: confusion matrix, accuracy, recall, precision, cross entropy, f1-score. The results are compared with these of the most representative image classifiers for the kind of classification we target, in terms of performance, as well as the size of the retrained frozen graph on our dataset. Such a classification could serve as a valuable tool in clinical medicine offering an automated, diagnostic tool for microbiome-related morbidities, thus allowing accurate early diagnosis and the design of personalized and targeted therapeutic approaches.