Artificial intelligence (AI) has potential to improve the diagnosis of melanoma, but the road to its widespread adoption requires overcoming various challenges, according to a new review.1
The promise of AI has been anticipated as traditional diagnostic methods, including physical examination and nodal assessments, have lacked reliability and face variability in interpretation based on who is reviewing.
Outside of patient diversity, the researchers call for variety in data types to strengthen AI models, with a compilation of patient records with genomic data and images. | Image credit: natali_mis – stock.adobe.com
“Given the growing incidence of melanoma in the world, applications based on AI can help reduce the burden on doctors, simplify the diagnostic algorithm, and provide populations with equal access to adequate treatment,” the researchers explained.
The review, published in International Journal of Intelligent Systems, explored the current landscape of AI models being used to aid in the diagnosis of melanoma. These approaches include machine learning, deep learning, and mixed-approach methods.
Newer approaches like convolutional neural networks (CNNs) have gained steam for their ability to detect features in images and videos, and in turn have been explored in health care, including for melanoma.2 In one study of more than 2600 images, the deep learning approach had an 88% accuracy rate in classifying skin lesions as malignant or nonmalignant. Another study has demonstrated a 96% accuracy rate, showing accuracy improvements compared with earlier AI methods like Support Vector Machines (SVMs).
SVMs have demonstrated the ability to detect melanoma early, with accuracy of up to and more than 87%. Data have shown that the model is able to use just 6 identifiable factors to make a diagnosis.
Some models have approached AI with a hybrid of CNN and SVM, combining CNN’s ability to automatically pinpoint important features of images without the need for manual input with SVM’s ability to classify data.
“According to the research, various AI models could perform equally well or even better than experts in dermatology in specific situations, since most of them reported accuracy, sensitivity, and specificity levels higher than 85%. Nevertheless, several difficulties make it hard to put these findings into practice,” explained the researchers, noting that methods like CNNs struggle with widespread uptake.
In addition to resourcing issues, with many health care organizations facing structural and financial challenges for implementation, the researchers highlighted inconsistencies in the performance of current AI models. Inconsistent results are driven by differences in how these models are built and the data that they are trained to assess. Current models often leave an absence of explanation of their results, highlighting an opportunity for improved models to offer a basis for their diagnosis.
Improvements in the data used for AI approaches were highlighted by the group. It’s well documented that clinical trials, including those for melanoma, are often not representative of true patient populations in the real world.3 As a result, lesion images included in studies remain limited for analysis. The researchers emphasized the need for data on a variety of patient types, including those with different skin types and with different stages of disease.
Outside of patient diversity, the researchers call for variety in data types to strengthen AI models, with a compilation of patient records with genomic data and images. Improvements in AI-based models also require health care organizations to work closely with researchers and developers of such technology to share data.
At the same time, as patient populations and cases evolve, AI models must adapt with it, balancing both flexibility and accuracy.
“AI should be applied in health care after being tested in studies that follow its performance and effects in various settings,” the authors concluded. “Improved user interfaces and immediate support for making decisions must be available for practical use by doctors. It will be vital for AI developers, doctors, and healthcare policymakers to cooperate in order to develop rules that allow safe, ethical, and effective implementation of AI technology for early diagnostics of melanoma.”
References
1. Alam F, Ullah A, Shah D, Ali S, Tahir M. Artificial intelligence in melanoma detection: A review of current technologies and future direction. Int J Intell Syst. Published online June 13, 2025. doi:10.1155/int/3164952
2. Kwiatkowska D, Kluska P, Reich A. Convolutional neural networks for the detection of malignant melanoma in dermoscopy images. Postepy Dermatol Alergol. 2021;38(3):412-420. doi:10.5114/ada.2021.107927
3. Donia M, Kimper-Karl ML, Hoyer KL, Bastholt L, Schmidt H, Svane IM. The majority of patients with metastatic melanoma are not represented in pivotal phase III immunotherapy trials. Eur J Cancer. 2017;74:89-95. doi:10.1016/j.ejca.2016.12.017