June 3, 2024
Dermatology Times
May recently concluded Melanoma and Skin Cancer Awareness Month, however, the importance of sun protection and early detection doesn’t end with May. Artificial intelligence (AI) continues to be a conversation in dermatology, especially concerning skin cancer detection. To better understand how certain AI platforms can assist dermatology clinicians in data collection versus taking on the role of detecting skin cancer, OM1’s Joseph Zabinski, PhD, MEM, spoke to Dermatology Times about the future role of AI. Zabinski is the vice president and head of commercial strategy and AI at OM1, a real-world data, AI, and technology company with a focus on chronic diseases.
Q&A With Joseph Zabinski, PhD, MEM
Dermatology Times: How have you seen recent trends and opinions change towards AI detection platforms for skin cancer?
Zabinski: The greatest change I’ve observed recently is less in the capabilities of these kinds of tools, and more in their broader acceptance by patients, and clinicians. This is not yet widespread by any means, but tools that analyze images to provide some risk information, including around potential skin cancer are becoming less ‘exotic’ – as they normalize, and improve, usage will continue to increase.
Overall, there seems to be more momentum in AI than the industry has seen in a while with a shift towards greater maturity and real-world applications. This focus on clinical care pathways and large-scale settings may lead to implementation of well-functioning AI in clinical settings, especially to surface ‘hidden’ or lost patients for diagnosis and access to effective treatment.
Dermatology Times: Where do AI skin detection platforms show success? Where are they still falling short?
Zabinski: These platforms are generally most successful in lowering the barriers patients face to getting some insight into dermatologic issues they’re facing. Even if performance isn’t perfect, being able to take a picture at home and add some context through AI is a powerful augmentation to patients’ experience. That said, these platforms are still not perfect; they’re not always trained on representative datasets, don’t necessarily generalize across patient populations, and are not always effective at communicating uncertainty or variability in their outputs.
Greater adoption also remains an obstacle and will only be overcome if we are clear on the value AI creates to build trust among patient populations. This includes seamlessly integrating technology into clinical workflows by efficiently identifying risks and informing personalized treatment plans.
Dermatology Times: What role does OM1 play in the education of ethical AI use, especially in dermatology?
Zabinski: There’s a lot of important work to do in strengthening ethical AI use across clinical areas, for example, developing formal definitions of bias and implementing human-in-the-loop methods to focus on managing and minimizing flaws so that society can benefit from AI more equitably and with less risk.
At OM1, our main focus right now is emphasizing that these tools need to be trained on high-quality, representative datasets; that clinician input is critical to ensuring they make sense in clinical contexts; and most importantly, that AI-derived information supports – not supplants – patients’ and physicians’ decision-making. We believe strongly that AI tools built and used correctly can enhance patients’ informed consent in their care by personalizing insight about their outcomes.
Dermatology Times: Does OM1 have any current projects or models that are dedicated to skin cancer data collection and prediction?
Zabinski: We haven’t yet worked in skin cancer specifically but have experience in other areas of both skin disease and cancer (e.g., breast, colon). We’ve found that high-quality, longitudinal data in these areas can help differentiate patient journeys, inform risk assessments, and support early detection.
Dermatology Times: During Melanoma and Skin Cancer Awareness Month, what insights do you have for clinicians who are worried about AI platforms used for the detection of skin cancer and/or melanoma?
Zabinski: In my conversations with clinicians, 2 concerns typically come up around these kinds of uses of AI. First, they worry these tools will provide inaccurate information – especially important when that information makes it to the patient, and potentially creates either unnecessary anxiety from a false positive, or a sense of security from a false negative where treatment may be needed. Second, they’re concerned that patients will be deluged with information that they don’t really understand, and that will spill over to the clinicians when patients begin asking them to interpret all sorts of AI-derived data. In both cases, my advice is to focus attention on where AI improves current practice, in ways that minimally disrupt current practice. Good and bad tools exist, and focusing on where good can be done helps alleviate anxieties, builds trust, and improves patient care.
Dermatology Times: Do you have any closing thoughts about skin cancer and AI?
Zabinski: As with other forms of clinical AI, tools to detect cancer from skin images are going to get more popular, more accurate, and more helpful. There are challenges to navigate along the way, but the eventual impact on patient outcomes will be substantial, including improving early detection rates, establishing timely interventions, and personalizing treatment plans.