How artificial intelligence is transforming pigmentary dermatology—An overview

Introduction Artificial intelligence (AI) is reshaping dermatology, including the realm of pigmentary disorders. By integrating machine learning algorithms and image analysis, AI is capable of enhancing diagnostic accuracy, personalized treatment plans, and improved patient outcomes. The key role of AI in pigmentary dermatology is image analysis. It depends on high-quality image acquisition and pre-processing to remove artifacts and enhance lesion clarity for accurate segmentation and border detection. This is useful in diagnosing difficult lesions like melasma, post-inflammatory hyperpigmentation, and vitiligo, where low contrast and ill-defined margins make visual assessment difficult. Beyond diagnosis, AI objectifies findings by quantifying features such as pigmentation patterns, color uniformity, and texture. Apart from benign pigmentary lesions, AI is a game-changer for the detection of difficult melanomas. Quantification of various features studied in the algorithm for melanoma assessment is based on dermoscopic (ABCD or pattern analysis) or clinical findings (ABCDE).[1] AI algorithms analyze color histograms, asymmetry, and textural cues to flag suspicious moles with impressive accuracy. The factors that can influence the analytic power include a high variation of images; anatomical location of the lesion; subject variables, like age; imaging parameters, like angle of light or camera; and the direction from which the lesion image is photographed or seen. Most commercially available AI-based tools like DANAOS, DBDermo-Mips, and MoleAnalyser and apps such as SkinVision and MoleScope use dermoscopy for early detection and monitoring of pigmented skin lesions, especially skin cancers.[1] While helpful for screening, apps like SkinVision and Molescope have several limitations including variable accuracy resulting in false positive or negative results, image quality issues, limited diagnostic scope, and lack of clinical context regarding patient’s medical or family history. Thus, these tools should supplement—not replace—regular dermatological check-ups. Currently, AI models are based on images collected mostly from the skin lesions of light-skinned individuals. These image banks lack representation of the skin of color (SOC) images, thereby posing challenges when it comes to AI-assisted image analysis of lesions in such cases. Some researchers have proposed training AI models to recognize “darkened” images in datasets to simulate SOC patients. The fast contrastive unpaired translation is one such technique that has been used to “darken” the color of the skin in the images without altering the basic pathological skin lesion. Although such techniques can serve the purpose of AI-assisted differentiation of melanoma from non-melanoma skin cancers as noted in some studies, darkening the skin images may not be the most effective method to teach AI models to evaluate SOC dermatosis. Instead, it will be prudent to incorporate more images of SOC lesions from clinics to the datasets to increase representation.[2] As an alternative to Fitzpatrick scale, the Monk Skin Tone (MST) scale is a validated scale being used to train AI datasets for its more inclusive representation of skin tones with 10 categories and is not solely based on sun sensitivity, thus reducing bias and addressing issues like regional variation and global reach. The MST scale and its dataset are openly available and their inclusion can ensure more robust and equitable AI systems.[3] Darker skinned individuals are more prone to develop pigmentary diseases and melasma that constitutes one of the major causes of pigmentation, and its incorrect diagnosis hampers treatment. Accurate diagnosis is, therefore, essential and AI models are increasingly taking this role. Deep-learning AI models are being employed to diagnose melasma or post-inflammatory hyperpigmentation. High-performance models like DenseNet, ResNet, Swin Transformer, and MobileNet can assist in characterizing hyperpigmentation due to melasma and non-melasma conditions. These show excellent performance with clinical images taken by VISIA software, Canfield Scientific Inc, USA an AI-generated platform that can increase their efficiency by employing multiple view modes in images superimposed with UV spots, red areas, brown spots, and porphyrins, which may otherwise hamper AI-assisted diagnosis of pigmentary disorders of the face.[4] AI tools are also being used to predict the development of post-inflammatory hyperpigmentation, especially after elective procedures like lasers, and hence are aiding in improving patient and procedural outcomes.[5] Quantification of melanin amount may help in guiding treatment response in disorders like melasma. A proposed combination of full-field optical coherence tomography (FF-OCT) helps to assess melanin features of melasma lesions and perilesional skin of the cheeks and a computer-aided detection (CADe) system may help to quantify the melanin. This system combines spatial convolutional neural networks (SC-DnCNN) and through various visual processor techniques, melanin characteristics like area, distribution, intensity, and shape can be assessed. Comparing the differences between the lesion and perilesional skin, confetti melanin in stratum spinosum and a grain melanin at the dermal–epidermal junction help to characterize the depth of pigmentation which may aid in choosing the appropriate treatment modality. The CADe system with SC-DnCNN tool thus helps in assessing the area of distribution, depth, and configuration of melanin on FF-OCT images.[6] Further scope of AI in pigmentary disorders may be found in sequential digital dermoscopy (SDD) and automated mapping of pigmented skin lesions extracted from total body scanning systems. Time-consuming tests like nevus counting to stratify at-risk patients can be undertaken by AI tools from three-dimensional total body photography. The evaluation of sequential dermoscopic images can be tedious and can be plagued by legal and safety concerns as a result of an inability to detect changes in SDD manually. Automated algorithms could help by detecting these minor changes that can lead to early detection of carcinomatous changes in nevus.[7] Although the uses and benefits of AI are multiple and ever-evolving, some limitations of AI remain a concern. AI tools are entirely dependent on the quantity of data on which they are trained. Unless there is a huge data set available, deep learning AI tools cannot perform the required tasks of accurate diagnosis owing to huge variations in the expression of any disease on the human skin, for instance differentiating between benign nevi or more sinister diseases like melanoma skin cancers.[1] There is still a lack of standardization of image quality assessment. Many studies also note the absence of considering clinical factors while diagnosing like age, sex, location, actinic damage, and occupation. Dermatologists perform total body examination which may shed light on the diagnosis, subtle clues like the macroscopic Ugly Duckling sign (an atypical nevus amongst several other typical nevi), the Little Red Riding Hood sign (a benign looking nevi but is different from the rest of the nevi on the body), and dermoscopic nevoid patterns. This approach enhances the accuracy, sensitivity, specificity, and warrants for a more holistic outlook while diagnosing pigmentary conditions.[7] AI can significantly impact clinical practice by using its algorithm to analyze images and clinical data that aid in the diagnosis of pigmentary disorders, help tailor personalized treatment plans to patients based on unique skin characteristics, and enhance patient monitoring by tracking the changes over time. AI will aid in research work by the identification of disease patterns and trends in pigmentary disorders and AI algorithms can be trained to predict the likelihood of developing diseases based on genetic and environmental patterns. Also by analyzing large datasets, AI can identify new therapeutic targets and reduce healthcare outcome disparities by ensuring inclusivity in research and treatment. The application of AI in pigmentary disorders is riddled with several ethical and “black box” concerns such as bias from non-diverse datasets, data privacy, and lack of transparency in decision-making. Misclassification and recommending inappropriate therapies can erode trust. These issues can be mitigated through diverse datasets, explainable AI, and human oversight. When integrated with clinical workflows and EHRs (electronic health records), AI can enhance accuracy, streamline tasks, and support clinicians in managing complex cases. Integrating AI systems with clinical decision-making will provide dermatologists with accurate and effective decisions by leveraging the strength of both of them, and integration with electronic health records will help streamline clinical workflows. Acceptability and consumer trust in AI is another barrier to the implementation of AI in clinical practice. Human-to-human relationships in doctor–patient interaction cannot be underestimated. However, the presence of AI in healthcare and its use when guided by protocols will usher in a new smart era in healthcare. We need to prioritize the inclusion of AI in a standardized manner in our activities to provide optimum benefits to the patients.[7] The inclusion of AI in the medical curriculum should be considered to strengthen diagnostic and problem-solving skills, for individual learning, and hasten evaluation procedures. In the very near future, medical students will need to learn this new approach to education and incorporate technologies like AI and machine learning in their practice and research and aid further in revolutionizing healthcare.[8] Financial support and sponsorship Nil. Conflicts of interest There are no conflicts of interest.