This year, we had the opportunity to attend the American Academy of Dermatology’s Virtual Meeting Experience. It was a great experience despite the conference being entirely virtual. We had the opportunity to listen to pre-recorded lectures, as well as interact with live panels in each session via real-time chat functions. Among the numerous outstanding presentations, there was one session we found especially interesting: “Putting Technology to Work for You and Your Patients: Augmented Intelligence and its Role in Enhancing the Practice of Dermatology.”
A knee-jerk reaction to artificial intelligence (AI) is the concern of rendering physicians obsolete. Over the past few years, the volume of research on AI and machine learning has rapidly grown. Claims that AI performs superior to doctors have generated plenty of hype calling for accelerated implementation of AI tools in practice. Despite this hype, the panel presented a variety of concerns, including limitations of studies, issues with data privacy, and the risk of increasing care inequality. Ultimately, the future of AI lies in augmented intelligence (AuI) –– the concept that AI will serve an assistive role to enhance the patient-physician relationship rather than replace it.
AuI has several use-cases that are being explored today. Dr. Bui, from Google Health, highlighted a model that may help triage patients to prioritize urgent cases, and Dr. Lee referenced a model capable of identifying suspicious pigmented skin lesions using images from consumer-grade cameras. AuI can even be used to detect poor quality images and guide patients to take better photos, thereby optimizing telehealth workflows. Dr. Lee’s presentation also explored the potential to augment patient experiences, improve access to underserved populations, refine clinical efficiency and provider engagement, and optimize resource allocation. A majority of patients and dermatologists believe AI will have a positive impact on care, especially if AI and humans act in symbiosis.
However, it is easy to get lost in the excitement of this rapidly evolving technology. While it can improve access to care, AuI also has the potential to exacerbate inequalities. Dr. Adamson gave the example of an AI triage software that used cost as a proxy of medical need. Since Black patients are historically allocated fewer resources, the algorithm drastically underestimated the number of Black patients needing additional care. Inadequate representation of certain populations will result in algorithms that perpetuate biases. Machine learning algorithms are only as good as their training, so we must evaluate these new tools with the same vigilance used to evaluate other medical interventions. Given AI is still in its early stages, we have the opportunity to minimize — and even prevent — biases and disparities that may arise from AI.
A challenge involved in developing AuI algorithms is the need for large and diverse datasets. We agree with Dr. Rotemberg’s call to action for standardizing clinical imaging and data labeling. Standardizing image quality (e.g., lighting, distance, file size, clarity) and labeling (e.g., patient info, anatomic site, diagnosis) will be essential for building robust databases. Furthermore, it is especially important to ensure diversity in the pictures we take. The skin of patients of color is already underrepresented in dermatology education. Including these patients, perhaps even oversampling them, may reduce the risk of biased algorithms. In addition, we have an obligation to patients to maintain their autonomy and be transparent in collecting and sharing their data. This includes getting informed consent, ensuring data is stored safely, and ensuring data, is used for its intended use.
There is no doubt AI, and especially AuI, will become a part of medical practice in the future. However, the way this technology is implemented and the potential benefits it can provide are dependent on how we embrace it. The private sector will continue to make breakthroughs in AI technology; however, clinicians should be leading the way to guarantee the best outcomes for our patients. We know what our practices need and what our patients need. The more we know, the better we can inform patients about AI-enabled products and services. We hope others are as excited as we are to help AuI augment patient-physician relationships and enhance the care we provide. Thank you to Drs. Adewole Adamson, Caroline Nelson, Carrie Kovarik, Ivy Lee, Veronica Rotemberg, Susan Huang, Peggy Bui, and Justin Ko for the great discussion about this promising technology. We look forward to hearing more about progress in AI at the next AAD Meeting (hopefully in person)!
Image: Andrey_Popov / shutterstock
