PhD Scientific Days 2021

Budapest, 7-8 July 2021

CL_V_L: Clinical Medicine V. Lectures

Development of an automated, artificial intelligence based system to recognise, diagnose and follow up nail symptoms and disorders

Dorottya Keresztes1, Réka Kovács1, Júlia Liza Szebényi1, Péter Seffer2, Tamás Szépe2, Vilmos Bilicki2, Rolland Péter Gyulai1
1 University of Pécs Department of Dermatology, Venereology and Oncodermatology
2 University of Szeged Faculty of Science and Informatics

Text of the abstract

Introduction: The differential diagnosis of nail disorders and the assessment of nail disease severity is tedious work, and requires special expertise. Furthermore, nail severity scores do not always correlate with the true severity of symptoms.
Aim: Our aim is to develop an automated, artificial intelligence based system to recognise, diagnose, score and follow up nail symptoms and disorders.
Methods: We have developed a mobile phone based nail examiner device for the standardized photo documentation of nail symptoms, equipped with a custom 3D printed setup, special optics and lighting. It captures 16 images for each nail in a standardized automated way, and stores them along with cloud-based documentation technology. For each nail a task is created for annotation, images are later annotated, and analyzed using artificial intelligence. Patients seen for different dermatologic and nail disorders at the Department of Dermatology, Venereology and Oncodermatology of University Pécs were enrolled in this investigation. Nail images and detailed medical history were collected. In case of onychomycosis, nails were subjected to mycologic analysis including native KOH preparation and/or culture.
Results: Since January 2019, 2283 patients were enrolled, and 3417 nails were documented. So far we have annotated 2599 tasks containing a variety of nail diseases. In psoriatic patients the most common symptoms were: pitting (43%), longitudinal ridging (42%) and splinter hemorrhages (40%). In patients with filamentous fungal infections the most common symptoms were: yellow nail (48%), hyperkeratosis (46%), and white nail (26%); in patients with sprouting fungi these were onycholysis (100%), red discoloration (100%), and hyperkeratosis (92%). Establishment of the incidence and extent of nail symptoms was made, and correlated with specific scoring systems (NAPSI, Target NAPSI, OSI). We found that these systems do not correlate well enough with the severity of the symptomes.
Conclusion: We have developed an innovative nail examiner device, and established one of the largest annotated nail image database so far. By using deep neural network based artificial intelligence, we are currently developing a system capable of recognizing specific nail symptoms with high specificity and sensitivity.
Funding: GINOP-2.1.2-8-1-4-16-2017-00002

University and Doctoral School

University of Pécs Medical School, Doctoral School of Clinical Medical Sciences