In a pioneering study published in Lancet Digital Health, mobile phone-powered artificial intelligence (AI) has emerged as a formidable rival to specialists in the diagnosis of pigmented skin cancer. This report explores whether AI can effectively support the diagnosis and management of skin cancer, demonstrating that it not only outperforms junior doctors but also matches the accuracy of seasoned specialists in clinical settings.
Mobile AI surpasses junior doctors and specialists
In an endeavor aimed at transforming the landscape of skin cancer diagnosis, a recent multinational clinical trial brought together clinicians from the Sydney Melanoma Diagnostic Centre in Australia and the Division of Dermatology at Austria’s Medical University of Vienna. These clinicians, classified as either specialists or novices, were entrusted with the task of evaluating suspicious pigmented skin lesions.
The study focused on adult patients with skin types falling within the modified Fitzpatrick I-III range and lesions larger than three millimeters in diameter. Subsequently, each clinician’s diagnosis was compared to an assessment made by artificial intelligence (AI).
The study, which scrutinized 172 suspicious pigmented lesions from 124 patients, revealed intriguing results. The AI, equipped with a dermoscopy phone attachment, surpassed the performance of novice doctors and demonstrated comparable accuracy to specialists. This mobile AI, based on a seven-class algorithm, exhibited superior diagnostic capabilities, achieving results on par with seasoned experts in the field. In terms of diagnostic accuracy, the seven-class AI algorithm equaled that of the specialists while outperforming the novices.
Comparing mobile AI to existing algorithms
Significantly, it is worth noting that the artificial intelligence (AI) system developed for this research endeavor not only exceeded the performance of pre-existing ISIC AI algorithms but also stood as a testament to its exceptional robustness, which can be attributed to its extensive and diversified training dataset. The remarkable superiority in the performance of the seven-class AI system is, in fact, a direct consequence of the rich and diverse dataset it was trained on, rendering it a truly formidable contender in the realm of skin cancer diagnosis.
But, it is important to underscore that when it came to making crucial management decisions, the mobile AI system exhibited a noticeable shortfall when compared to both seasoned specialists and relative novices in the field.
Also, the primary focal point of this study, namely the balanced multiclass accuracy, remained consistently high throughout, effectively dispelling any lingering concerns pertaining to potential overfitting issues.Also, despite initial reservations regarding the performance of the ISIC AI algorithm when presented with images from untrained sources, the seven-class AI system consistently upheld its diagnostic accuracy, reassuring the scientific community about its reliability in diverse clinical contexts.
No compromise on accuracy
This clinical trial illuminates the considerable promise inherent in a mobile AI-driven system designed for the diagnosis of skin cancer via the analysis of dermoscopy images. In stark contrast to prior investigations, which necessitated the utilization of prohibitively costly hardware, this research unequivocally underscores the viability of a straightforward mobile phone technology as a means to achieve both precise and cost-effective skin cancer diagnosis. By effectively closing the divide that traditionally separates specialists from individuals lacking expertise in dermatology, mobile AI technology holds the potential to substantially enhance the accessibility and efficiency of skin cancer diagnosis within the clinical domain.
