VCOM-Carolinas Spartanburg, South Carolina, United States
Introduction/Purpose: The American Academy of Dermatology and the American Cancer Society report that melanoma has the highest mortality rate of all skin cancers and the overall incidence rate of melanoma continues to rise each year. Melanoma is highly treatable in the early stages before spread, but five year survival drastically declines to 68% and 30% with regional and distant metastasis, respectively. Melanoma has several characteristics allowing for visual recognition including asymmetry, irregular borders, multiple colors, and large diameter (greater than 6 mm). Diagnosis of melanoma is predominantly made by tissue biopsy which is viewed under a microscope by board-certified histopathologists. Software systems fueled by Artificial Intelligence, including Convolutional Neural Networks and Deep Learning Systems, have developed a level of performance acceptable for application in the field of dermatology. The success demonstrated by these systems stands to enhance techniques for image recognition and improve efficiency in the diagnostic domain.
Methods or Case Description: A systematic review of the literature was performed with pre-determined exclusion criteria to evaluate evidence for the use of AI in the recognition of malignant melanoma. Studies exploring the performance of melanoma recognition of Whole-Slide Images (both analog and digital) by AI programs compared to pathologists were analyzed. Performance outcomes were measured using accuracy (with respect to the gold standard reference of classification by expert dermatopathologists), sensitivity, specificity, and area under the curve. Some studies further evaluated results using Cohen’s kappa.
Outcomes: Results across all studies show that AI systems, as they currently stand, have demonstrated the ability to identify melanoma at a high-accuracy level of classification similar to or greater than that of pathologists. The achievement of these systems indicates that AI can be a valuable tool to aid clinicians and may reduce inter-observer inconsistency. The findings of this review demonstrate that the use of AI can, in the smallest degree, provide a triage to pathologists and will be instrumental in helping to improve the overall accuracy of melanoma recognition. When analyzing the measures-of-validity results for each article, it was observed that most studies had a sensitivity above 87% and higher than pathologists in all cases. High sensitivity is vital to melanoma screening by way of reducing false negatives which can result in delayed diagnosis and treatment. Melanoma is highly treatable in the early stages, but survival rates decline drastically after metastasis occurs. Implementation of AI with high sensitivity is necessary to expedite diagnosis, acquire biopsy or further testing, and improve patient outcomes. Further, AI may be beneficial on a broader scale by allowing primary care physicians, who provide the vast majority of healthcare to rural and underserved populations, to achieve an equally accurate and efficient diagnosis of melanoma. The use of AI, under the guidance of primary care physicians, can be efficiently utilized as a screening and detection mechanism for biopsy, the gold standard of melanoma diagnosis. Teledermatology referrals have proven to be effective in reducing the average time to biopsy. AI offers the possibility of massive reduction in the biopsy-to-results turnaround time that will increase early diagnoses and improve the survival rate of patients with melanoma.
Conclusion:
AI systems have demonstrated the ability to identify melanoma at a high-accuracy level of classification similar to or greater than that of pathologists
The use of AI can, in the smallest degree, provide a triage to pathologists and will be instrumental in helping to improve the overall accuracy of melanoma recognition
We propose that AI may be beneficial in allowing primary care physicians providing healthcare to rural and underserved populations to achieve an equally accurate and efficient diagnosis of melanoma
AI offers the possibility of massive reduction in the biopsy-to-results turnaround time that will increase early diagnoses and improve the survival rate of patients with melanoma
