End-to-end web application for image analysis using neural networks
Analyzing large sets of image data takes a lot of effort and is still a manual process that requires expert knowledge, especially with biomedical images. Convolutional neural networks (CNNs) have an enormous potential in the medical field due to their ability to extract important features automatically. To make this technology available to a broad spectrum of end users, Fraunhofer researchers have developed the intuitive, web-based end-to-end pipeline MyDeepLearn: It supports users almost from the beginning with a partly automated, easy-to-access pipeline with high usability that reduces the amount of time and manual work. The end-to-end web application enables, for example, detection of skin alterations in melanoma images and classification of skin cancer using neural networks.
Valuable support for early diagnosis
MyDeepLearn links the CNNs in the back end with the web application in the front end. It provides users with a vector-based editing tool to post-process predictions or generate annotations, in addition to visualizing the data and providing different evaluation methods. It helps researchers and physicians to evaluate large sets of image data and supports early diagnosis – in particular in situations where there are only few annotated data available, as is often the case in the medical domain.
The web application can improve the CNNs iteratively with an interactive "human-in-the-loop" approach, which means that the user is involved in the evaluation and can actively change it. The workflow thus provides users with constantly improving technology, while continually expanding the data set to make this technology even more accurate.