Recognizing Common Skin Diseases in the Philippines Using Image Processing and Machine Learning Classification

Artificial Intelligence on Diagnostic Aid of Leprosy: A Systematic Literature Review

“…Binary classification occurred in 43% of the selected articles [ 12 , 14 , 76 , 82 , 83 , 87 , 94 , 95 , 96 ]. The remaining 57% of the articles used multiclass classification tasks, in which the models classified different skin diseases and, among them, leprosy [ 84 , 85 , 86 , 88 , 89 , 90 , 91 , 92 , 93 , 97 , 98 , 99 ]. Also, results revealed that 24% were classified as leprosy or not.…”

“…The data types used in most models were images of skin lesions, and models classified leprosy against other dermatological diseases [ 12 , 82 , 83 , 84 , 85 , 86 , 88 , 89 , 90 , 91 , 92 , 93 , 97 , 98 ]. However, Barbieri et al [ 12 ] used images of skin lesions combined with clinical information to develop an AI model.…”

“…Some authors used numerical data, so requiring the data to be normalized [ 12 ]. Researchers who used datasets with images and ML classical algorithms in [ 86 , 89 , 92 , 93 , 97 , 98 , 99 ] had to apply various image preparation techniques and feature extraction. The features most explored by the authors were related to texture and edges, with applications of spatial filters aiming to correct, smooth, or enhance specific regions.…”

“…The features most explored by the authors were related to texture and edges, with applications of spatial filters aiming to correct, smooth, or enhance specific regions. In addition, image compression techniques (e.g., YCbCr algorithm [ 97 ] and DCT [ 89 , 98 ]), segmentation-related techniques (e.g., binary mask [ 92 , 97 ], histogram [ 90 , 97 ], OTSU [ 86 ], global thresholding [ 93 ]), and image noise reduction techniques (e.g., median filter, smooth filter [ 92 , 97 ]) were explored by the studies. Jin et al [ 90 ] used the ResNet-50 and VGG16 architectures and the HOG technique for feature extraction.…”

“…Researchers in [ 12 , 14 , 76 , 87 , 94 ] used the classical algorithms MLP, LR, RF, and DT to train numerical datasets. Researchers in [ 86 , 89 , 90 , 92 , 93 , 97 , 98 , 99 ] utilized classic ML algorithms DT, SVM, FFBPN, and ANN to build models with image datasets. Researchers in [ 82 , 83 , 84 , 85 , 88 , 91 , 95 ] used the MobileNet-V2, DenseNet-121, Inception-V3, ResNet-50, EfficientNet-B2, LeprosyNet, and Siamese Network architectures to develop models with image datasets.…”