In a study spearheaded by the Stanford Center for Digital Health, it has been found that employing artificial intelligence algorithms rooted in deep learning significantly enhances the precision of skin cancer diagnosis for medical professionals, including doctors, nurse practitioners, and medical students. According to a recent report from Stanford Medicine, these computer algorithms, driven by AI technology, offer substantial assistance in diagnosing skin cancers with greater accuracy, benefiting healthcare providers across various specialties. Even dermatologists, while experiencing a comparatively lesser enhancement in diagnostic accuracy, find value in the guidance provided by AI. Professor Eleni Linos, MD, who leads the Stanford Center for Digital Health, emphasizes the collaborative potential of AI alongside physicians in improving patient care. The study, published in npj Digital Medicine on April 9, underscores the potential of AI to address critical healthcare challenges by fostering interdisciplinary collaboration among fields such as engineering, computer science, medicine, and the humanities. Senior author Professor Eleni Linos, MD, alongside lead authors Dr. Jiyeong Kim and Dr. Isabelle Krakowski, highlights the promising implications of their research in advancing medical diagnostics.

Kim highlighted that previous research has primarily concentrated on evaluating the performance of AI in comparison to that of physicians. However, our study took a different approach by comparing the diagnostic accuracy of physicians operating without AI assistance to those utilizing AI in the diagnosis of skin cancers.

In clinical settings, AI algorithms are increasingly being employed, particularly in dermatology. These algorithms are developed by inputting hundreds of thousands, or even millions, of images of skin conditions into a computer. These images are labeled with pertinent information such as diagnosis and patient outcomes. Through a process known as deep learning, the computer gradually learns to identify distinctive patterns within the images that are associated with specific skin diseases, including various types of cancers.

Once trained, an algorithm generated by the computer can analyze an image of a patient’s skin and suggest potential diagnoses, even when it hasn't encountered that specific image before.

However, it's important to note that these diagnostic algorithms are not standalone tools. They are utilized under the supervision of clinicians who evaluate the patient independently. Clinicians form their own conclusions regarding the patient's diagnosis and decide whether to accept the algorithm's suggestion.

An accuracy boost emerged when Kim and Linos’ team conducted a comprehensive review of 12 studies encompassing over 67,000 evaluations of potential skin cancers by various practitioners, both with and without AI assistance. The results showed that healthcare practitioners, operating without aid from artificial intelligence, achieved an accuracy rate of diagnosing about 75% of individuals with skin cancer, a metric known as sensitivity. Conversely, they correctly identified around 81.5% of individuals with cancer-like skin conditions but without cancer, a measure known as specificity.

Healthcare practitioners who utilized AI to assist their diagnoses demonstrated improved performance. Their diagnoses showed approximately 81.1% sensitivity and 86.1% specificity. While the improvement may appear marginal, it holds critical significance for individuals either erroneously informed they are cancer-free or those with cancer but inaccurately deemed healthy.

Upon segregating the healthcare practitioners by specialty or level of training, the researchers observed that medical students, nurse practitioners, and primary care physicians benefited the most from AI guidance, exhibiting an average improvement of about 13 points in sensitivity and 11 points in specificity. Although dermatologists and dermatology residents displayed superior overall performance, their diagnostic sensitivity and specificity also saw enhancements with AI.

Linos expressed surprise at the universal improvement in accuracy with AI assistance, regardless of the practitioner’s training level, fostering optimism regarding AI's integration into clinical care. The researchers at the Stanford Center for Digital Health, including Kim, aim to delve deeper into the prospects and obstacles surrounding AI-based tools in healthcare. They plan to investigate how physicians’ and patients’ perceptions and attitudes toward AI will shape its integration.

Understanding how humans interact with and utilize AI in clinical decision-making is a primary focus. Previous studies have highlighted factors such as a clinician’s confidence in their own judgment, the AI’s confidence level, and the agreement between clinician and AI diagnoses as influential in whether the clinician incorporates the algorithm’s advice.

Specialties like dermatology and radiology, heavily reliant on image-based diagnoses, are seen as ripe for AI integration. However, even symptom-based specialties and predictive modeling are anticipated to benefit from AI intervention. Beyond patient outcomes, the potential for AI to enhance diagnostic accuracy while saving time for doctors could alleviate physician burnout and enhance doctor-patient relationships.

Linos emphasized the inevitability of AI assistance in all medical specialties and stressed the importance of ensuring its equitable implementation to benefit patients across diverse backgrounds while supporting physician well-being. Contributions to the research came from researchers at the Karolinska Institute, Karolinska University Hospital, and the University of Nicosia, with funding from various entities including the National Institutes of Health, Radiumhemmet Research, the Swedish Cancer Society, and the Swedish Research Council.