Skin is the largest organ in the body, and as such there is a wide range of diseases and lesions that need to be treated. One of them is skin cancer. If a concern arises for a certain disease, a dermatologist takes a skin sample which is passed to a pathologist for further examination and diagnosis. Pathologists look at digital skin samples and search for indicators and risk factors. However, these scans are very large, so the diagnosis is time consuming and the process is very tedious. In many cases pathologists examine only a portion of the sample and consequently diseases might go undetected or be diagnosed incorrectly.
Our project’s purpose is the analysis of histopathological images of the skin - the skin samples scans. We aim to create an informative and convenient visualization tool to save time, simplify the process, and help focus the pathologists' efforts towards diagnosis. Our focus is on the epidermis – the top layer of the skin since detection of diseases is mostly carried out there. In addition, the automation will allow more detailed and accurate results, by reducing the tediousness of the current process.
The project is based on image processing algorithms, computer vision and artificial intelligence (AI) which rely on public medical databases and on previous research in the field. The algorithm segments the epidermis, extracts geometric properties, locates cells in the layer and classifies them into different types, focusing on cell types whose distribution has critical implications. We then extracted statistical and geometric information of the sample, visualized to the physician conveniently in a user interface.
The result was a basic tool that allows a pathologist to select a scan from an existing database, and perform an automatic analysis of segmentation into layers, marking the cells and classifying them into two types of cells with reasonable accuracy, which helps diagnose melanoma better.
Dr. Mati Rosenblatt, M.D