A new approach for vaginal microbial micrograph classification using convolutional neural network combined with decision-making tree (CNN-DMT)

Vaginitis, the most common disease of female genital tract infections, mainly relies on the morphological detection of the vaginal micro-ecological system to diagnose under the microscope. It affects women's normal life seriously, even their fertility. Since the morphological detection is very dependent on the experience of the observer, while the experienced doctors are mostly concentrated in large cities, the problem of diagnosis of vaginitis in rural women is extremely serious. Convolutional neural network (CNN), the typical algorithm of artificial intelligence, has shown great potential in many visual classification tasks. However, it is difficult to apply CNN method directly to the diagnosis of vaginitis. To solve the problem, this paper proposes an algorithm combining CNN with decision-making tree (CNNDMT) based on medical expert consensus. In a way of incorporating features automatically extracted by the machine and expert knowledge, automatic diagnosis of vaginitis disease is realized. Experimental results show that the CNN-DMT approach improves test accuracy by 8.46% over the leading CNN method, while enhancing the accuracy of normal bacterial flora by more than 15%.