Examine the Influence of AI Tools on the Workflow of Dermatology Practice, Focusing on Diagnostic Efficiency, Patient Management, and Clinician Decision-Making

Abstract

Introduction: Skin conditions are an important health issue. On average, individuals experience 1.6 skin conditions each year, with skin-related doctor appointments making up 20% of all primary care visits, of which approximately 35% are directed to a dermatologist. Machine learning (ML) models have the potential to assist primary care professionals by analyzing and enhancing intricate datasets. Furthermore, ML models are being more commonly used in dermatology for aiding in diagnosis through image analysis, particularly for detecting and categorizing skin cancer.

Aim: This research seeks to validate a machine learning image analysis model prospectively as a diagnostic aid for diagnosing dermatological conditions.

Method: In this upcoming study, 100 successive patients seeking care for a skin issue from a participating general practitioner (GP) in central Catalonia were selected. The planned duration for data collection was set at 7 months. Anonymized images of skin conditions were captured and fed into the ML model interface (able to detect 44 different skin conditions), which provided the top 5 diagnoses with the highest probability. The identical picture was also transmitted for a teledermatology consultation in accordance with the established workflow. The GP, ML model, and dermatologist's evaluations will be compared to determine the precision, sensitivity, specificity, and accuracy of the ML model. Each type of skin disease class will have its results displayed globally and individually through a confusion matrix and the one-versus-all approach. The amount of time needed to conduct the diagnosis will also be factored in.

Results: Patient enrollment started in June 2021 and continued for a duration of 5 months. At present, all participants have been enrolled and the images have been presented to the GPs and dermatologists. The examination of the findings has already commenced.

Conclusion: This research will offer insights into the efficacy and constraints of ML models. External testing is necessary for controlling these diagnostic systems for the implementation of ML models in a primary care environment.