ARTIFICIAL INTELLIGENCE APPLIED TO DERMATOLOGICAL DIAGNOSIS: WHERE ARE WE? / INTELIGÊNCIA ARTIFICIAL APLICADA AO DIAGNÓSTICO DERMATOLÓGICO: EM QUE PONTO ESTAMOS?

• DOI: https://doi.org/10.22533/at.ed.1595312602012

• Palavras-chave: Dermatologia; Inteligência Artificial; Diagnóstico.

• Keywords: Dermatology; Artificial Intelligence; Diagnosis.

• Abstract: INTRODUCTION:Dermatology, due to its visual nature, has proven to be a particularly favorable field for the application of artificial intelligence (AI) in the diagnosis of skin lesions. Since 2017, advances in deep learning and neural networks have enabled algorithms to achieve performance comparable to that of specialists in recognizing neoplasms such as melanoma and basal cell carcinoma. Currently, AI is applied across multiple imaging modalities, including dermoscopy and clinical photography, using advanced tools such as EfficientNet, ResNet, and segmentation techniques (U-Net). New approaches aim to reduce biases and increase clinical confidence, although challenges remain, such as limited data representativeness and the need for robust clinical validation.OBJECTIVE:The objective of this study was to understand the application of AI in dermatological diagnosis, highlighting technological advances and challenges for its integration into clinical practice.METHODS:A narrative literature review with a qualitative and descriptive approach was conducted using the PubMed (MEDLINE) database. The following MeSH descriptors were used: (“Artificial Intelligence” OR “Machine Learning”) AND (“Skin Neoplasms” OR “Dermatology”) AND (“Diagnosis” OR “Image Interpretation, Computer-Assisted”). Articles published between 2015 and 2025, in English, with full text available and related to the use of AI in the diagnosis of skin lesions were included. After screening and applying exclusion criteria, 144 studies were selected and analyzed qualitatively regarding clinical applications, lesion types, and model performance.RESULTS:Artificial intelligence is widely applied in dermatology, particularly for melanoma and other pigmented lesions. Convolutional Neural Networks (CNNs), deep learning models such as ResNet and Inception, are the most frequently used and demonstrate high performance, ranging from 74% to 98%, comparable to or even exceeding that of specialists. Machine learning techniques such as Support Vector Machines (SVM) and Random Forest are also utilized but show lower performance due to the need for manual error correction. Practical applications of AI include software integrated with digital dermoscopy and total body photography, exemplified by MoleAnalyzer Pro and DEXI, as well as mobile applications such as SkinVision, which demonstrate good sensitivity (80%) and specificity (78%). However, AI application still requires robust clinical validation and faces challenges such as data bias, lack of standardization, and the need for diagnostic confirmation via biopsy and histopathological examination.CONCLUSION:Artificial intelligence has emerged as a valuable tool in dermatology, with advances in deep learning and neural networks enabling performance comparable to that of physicians. Its applications range from digital dermoscopy to mobile applications, expanding diagnostic support. Nevertheless, challenges such as limited representativeness in databases, lack of standardized guidelines, and the need for multicenter validation continue to limit its clinical implementation. Therefore, AI should be regarded as a complement to medical practice, capable of optimizing screening and accelerating clinical decision-making with greater safety.