Researchers designed a computer-aided diagnostic model to classify, diagnose, and provide treatment recommendations for patients with psoriasis.

There has been a significant uptick in AI applications over the last two decades, with many studies evaluating the use of advanced technology for diagnosing dermatologic conditions, according to authors of a recent study in Computers & Industrial Engineering.

“Many research studies have employed machine learning and deep learning methods in dermatological problems, evaluating the response to treatment, disease course, epidemiology, and patient experiences,” wrote PhD Candidate Abtin Ijadi Maghsoodi and colleagues.

Despite these dramatic advancements in AI and deep learning, the researchers identified a gap in the knowledge base, as no study has examined a computer-aided diagnosis (CAD) system for diagnosing and classifying psoriasis or recommending treatment.

To address this, the investigators developed a study to determine the viability of deep learning methods in the development of CAD systems with an ensemble convolutional neural network (CNN) to identify, classify, diagnose, and provide treatment recommendations in patients with psoriasis.

The network determines whether an image depicts psoriasis by applying a binary classification procedure during the first step of the process (ie, Is the image psoriasis? Or is the image not psoriasis?). The network then classifies the images depicting psoriasis into one of seven categories: plaque psoriasis, guttate psoriasis, erythrodermic psoriasis, inverse psoriasis, pustular psoriasis, nail psoriasis, and psoriatic arthritis.

To add treatment recommendations, the researchers implemented a multi-criteria decision making (MCDM) system into the diagnostic algorithm to recommend the optimal treatment protocol.

The researchers discovered that their novel, hybrid, web-based decision and diagnostic support system determined whether an image depicted psoriasis with 91.9% accuracy. Furthermore, the system categorized the psoriasis images into one of the seven outlined categories with an accuracy of 93.29%.

“The results suggest that this system can assist clinicians and dermatologists in the reliable and time-efficient classification of psoriatic skin images and in recommending optimal treatment options,” Maghsoodi and colleagues said.