Table 1.
Summary of literature on the use of machine learning in predicting dermatological outcomes.
| Study | Type of supervision | End point | Results | Software(s) utilized | Data used |
|---|---|---|---|---|---|
| Emam et al. (24) | Supervised | Risk of discontinuation of biologic | AUC for predicted risk of discontinuation due to: Any reason 0.95 Lack of efficacy 0.91 Adverse event 0.88 Other reasons 0.80 |
Generalized linear model, support vector machine, decision tree, random forest, gradient boosted trees, deep learning |
n = 681 psoriasis patients 13 clinically relevant features per patient |
| Wang et al. (25) | Semi-supervised | Risk of developing non-melanoma skin cancer | AUC 0.89 Sensitivity 83.1% Specificity 82.3% |
Convolutional neural network (deep learning) |
n = 9,494, 1,829 non-melanoma skin cancer patients, 7,665 random non-cancer controls 20 clinically relevant features per patient |
| Roffman et al. (26) | Supervised | Risk of developing non-melanoma skin cancer | AUC 0.81 Sensitivity 86.2% Specificity 62.7% |
Artificial neural network (deep learning) |
n = 462,630, 2,056 non-melanoma skin cancer patients, 460,574 non-cancer patients 13 clinically relevant features per patient |
| Khozeimeh et al. (27) | Supervised | Response to wart treatment method | Cryotherapy: AUC 0.902, accuracy 80% Immunotherapy: AUC 0.813, accuracy 98% |
Fuzzy logic and adaptive network-based fuzzy inference system (ANFIS) |
n = 180, 90 patients in cryotherapy group, 90 patients in immunotherapy group 7 clinically relevant features per patient in the cryotherapy group, 8 in the immunotherapy group |
| Tan et al. (28) | Supervised | Complexity of reconstructive surgery after periocular basal cell carcinoma excision | Naïve Bayesian Classifier: AUC 0.854 PPV 38.1% NPV 94.1% ADTree: AUC 0.835 PPV 31% NPV 97% |
Decision table, Bayesian, tree-based methods, multivariate logistic regression, nearest neighbor classifier, support vector machine |
n = 156 periocular BCC patients 7 clinically relevant features per patient |
| de Franciscis et al. (29) | Supervised | Risk of developing chronic venous ulcers in patients with chronic venous disease | CVU group level of risk 32.38 ± 7.19% Non-CVU group level of risk 8.34 ± 3.38% |
Fuzzy logic to stratify CVD patients into CVU and non-CVU groups |
n = 77, 40 patients with CVU, 37 patients without CVU 27 clinically relevant features |