Artificial Intelligence App Shows Promise in Detecting Skin Changes

Image Credit: © Toowongsa - stock.adobe.com

The integration of artificial intelligence (AI) in dermatology is rapidly evolving, with the hope to enhance the diagnosis and monitoring of skin conditions ranging from benign ailments like acne and eczema to serious concerns such as skin cancer.¹ With the increasing prevalence of skin changes and the rising public awareness regarding skin health, the demand for efficient and reliable tools to monitor these changes has become more critical than ever.² The advent of smartphone applications designed for skin checks reflects a consumer-driven interest in innovative solutions. However, these applications must meet specific criteria, including user-friendliness, accessibility, affordability, and scientific validity.³

A recent study explored the findings of the SkinChange.AI (SCAI) app, which utilizes a novel AI-assisted approach to identify skin changes over time. The study provides insights into the potential of such technologies in aiding dermatologists and other clinicians in their practice.⁴

Study Design and Methods

Conducted at the Danish Research Center for Skin Cancer, the pilot study aimed to assess the feasibility of the SCAI app in detecting simulated skin changes. The study involved 24 healthy adults, aged 19 to 62, with Fitzpatrick skin types I-III. Using a controlled environment, the researchers applied adhesive test spots of various colors (black, brown, and red) to the participants' backs and legs. The SCAI app captured images before and after the application of these spots, ensuring standardized conditions through a customized lighting setup.

The SCAI app features an interface for standardized image capture and an AI backend that facilitates image alignment and comparison. The app employs advanced AI algorithms for background identification and spot detection, allowing for comparisons between paired images.

Researchers stated the primary outcome measures included the app's sensitivity and specificity in identifying the applied test spots. Statistical analyses yielded descriptive data, including predictive values for the app’s performance.

Results

The results reported from the pilot study revealed that the SCAI app exhibited a sensitivity of 92.0% and specificity of 95.5%. Researchers said the positive predictive value (PPV) was noted at 38.0%, while the negative predictive value (NPV) reached 99.7%. This indicates that while the app successfully identified true skin changes, it also generated a notable number of false positives.

Researchers said a classification of results by anatomical location showed enhanced sensitivity on the back compared to the legs, attributed to differences in image capture distance and background complexity. Moreover, the study found the app demonstrated better performance in identifying darker test spots compared to brown ones, suggesting that color differentiation could be a vital factor in the app’s efficacy.

Discussion

The study found the SCAI app shows potential as a tool for improving the monitoring of skin changes, particularly in populations that may lack regular access to dermatological care. The ability of the app to provide high NPV means that it can reliably rule out non-changes, a critical aspect in clinical settings to prevent unnecessary anxiety and healthcare burdens associated with false positives.

However, researchers said the low PPV raises important considerations. They noted high rates of false positives can lead to overdiagnosis and an increased workload for dermatologists, as they will need to evaluate numerous flagged changes that may not require clinical intervention. Researchers said this balance between sensitivity and specificity is a common challenge in AI applications and needs addressing in future iterations of the app.

Implications for Clinical Practice

The study stated the incorporation of AI tools like the SCAI app in dermatology could enhance patient engagement in self-monitoring skin changes, especially in populations at risk for skin cancers. The app’s design allows for users to capture high-quality images while maintaining standardized conditions, which they noted is crucial for consistent monitoring over time.

Furthermore, as AI tools evolve, continuous training and validation of algorithms with diverse datasets will be necessary to improve the detection accuracy of various skin lesions beyond the study's simulated changes. The study stated that future studies should focus on real-world applicability, assessing the app's performance on actual clinical lesions and a broader demographic to ensure robustness across different skin types and conditions.

Conclusion

In conclusion, the study found the SCAI app demonstrates promising capabilities in detecting simulated skin changes with high sensitivity and specificity. While the app's potential as a clinical aid in dermatology is evident, researchers believe addressing the challenges of false positives and broadening its application to real clinical settings are imperative for its successful integration into everyday dermatological practice.

References

1. Liopyris K, Gregoriou S, Dias J, et al. Artificial intelligence in dermatology: Challenges and perspectives. Dermatol Ther (Heidelb). 2022;12(12):2637-2651. doi:10.1007/s13555-022-00833-8
2. Furriel BCRS, Oliveira BD, Prôa R, et al. Artificial intelligence for skin cancer detection and classification for clinical environment: a systematic review. Front Med (Lausanne). 2024;10:1305954. Published 2024 Jan 8. doi:10.3389/fmed.2023.1305954
3. Freeman K, Dinnes J, Chuchu N, et al. Algorithm based smartphone apps to assess risk of skin cancer in adults: systematic review of diagnostic accuracy studies [published correction appears in BMJ. 2020 Feb 25;368:m645. doi: 10.1136/bmj.m645]. BMJ. 2020;368:m127. Published 2020 Feb 10. doi:10.1136/bmj.m127
4. Grove GL, Reedtz G, Vangsgaard B, et al. Artificial intelligence smartphone application for detection of simulated skin changes: An in vivo pilot study. Skin Res Technol. doi: 10.1111/srt.70056