Site Editor
Soo Park, MD
Soo Park, MD
Posted: Thursday, February 1, 2024
Can artificial intelligence (AI) help clinical practitioners to better assess skin lesions? Perhaps, according to a recent study. Kristen M. Meiburger, PhD, of Politecnico di Torino, Italy, and colleagues investigated the impact of DermoCC-GAN, an AI-based color constancy algorithm, on the skin lesion diagnostic routine. They found its use improved color constancy, thus lessening the variability of dermatoscopic images and qualitatively benefiting practitioners. DermoCC-GAN stands for the Dermatological Color Constancy Generative Adversarial Network. These findings were published in Skin Research and Technology.
Under normal conditions, dermatoscopic image quality is affected by lighting conditions as well as operator experience and device calibration, the authors explained. The DermoCC-GAN algorithm “reduce[s] this variability by making images [acquired under unknown light] appear as if they were acquired under defined light conditions…usually perfectly white light.”
Three dermatologists with varying experience levels performed two assignments (unpaired and paired evaluation tasks) to gauge key parameters—perceived image quality, lesion diagnosis, and diagnosis confidence—and all performances seemed to improve with the AI-based algorithm. The starting data set included 150 dermatoscopic images about equally divided among actinic keratosis, basal cell carcinoma, keratosis-like lesions, melanoma, and nevi.
Specifically, the dermatologists found that images normalized by the AI-assisted strategy were of higher quality on average than the original images. Observing the corresponding normalized images appeared to allow the clinicians to “obtain additional and useful information [to] support them in the diagnostic phase,” wrote the team. “Especially for a novice clinician, the impact of normalization on clinical assessment is greatest when the normalized image is provided alongside the original, leading to an increase in overall classification performance” as well as an increased level of dermatologist confidence.
In closing, Dr. Meiburger and co-investigators acknowledged the importance of “extreme caution…. We always suggest the simultaneous analysis of original and normalized images.”
Disclosure: The study authors reported no conflicts of interest.
JNCCN Spotlights
