ABSTRACT
Dermatology is a subject that relies on morphological features and the majority of diagnoses are based on visual pattern recognition. As such, dermatology is exceedingly suitable for applying artificial intelligence (AI) image recognition capabilities for assisted diagnosis. Much of the machine learning implementation in the field of dermatology encompasses the binary classification of malignant melanoma versus benign nevi. In classifying benign melanoma from malignant melanoma, machine learning neural networks have shown to have excellent performance and even outperform practicing dermatologists. Machine learning has also been used to differentiate other skin conditions such as acne, atopic dermatitis, and psoriasis among others and has similarly shown to provide high sensitivity and specificity. In the last few years, non-image machine learning for both diagnosing and prognosis/severity of dermatological conditions has also been developed. Future research will likely consist of developing ways to improve the accuracy of the model such as with data augmentation and by the addition of clinical information such as age, gender, race, time of onset, location, evolution, and family history to the dermatological image being used to train the neural network. In the upcoming years, AI can prove to be a great asset for dermatologists in diagnosing, selecting treatment, and monitoring the progression of their patients.
