Machine Learning Enhances Laser Treatment Planning

Image Credit: © Kaikoro - stock.adobe.com

The assessment of skin characteristics such as Fitzpatrick skin type, hyperpigmentation, redness, and wrinkle severity is crucial in determining the appropriate laser therapy for patients. With the rising demand for laser procedures, accurate evaluation of these characteristics is essential.¹ According to the American Society for Dermatologic Surgery (ASDS) Consumer Survey, the percentage of consumers considering cosmetic procedures increased from 30% in 2013 to 70% in 2023, with laser treatments emerging as the most sought-after option.² However, the increasing popularity of these procedures has highlighted concerns regarding the quality of care, particularly in non-physician-operated settings such as medical spas.³

Challenges in Skin Assessment and Laser Procedures

In the US, the availability of trained dermatologists remains limited, with only 1 dermatologist per 29,000 citizens. This shortage has led to an increase in nonphysicians, including nurses, aestheticians, and cosmetologists, performing skin assessments and administering laser treatments.⁴ Medical spas, which often employ nonphysician providers, now outnumber physician-based cosmetic practices in 73% of major US cities. Unfortunately, studies have shown that medical spas exhibit poorer safety records and patient outcomes compared to physician-led practices. Inaccurate skin evaluations can lead to complications such as burns, scarring, infections, and even vision loss.⁵

The Role of AI in Dermatologic Assessments

To address the shortcomings in skin assessment accuracy and improve treatment outcomes, researchers have explored the potential of artificial intelligence (AI) and machine learning in dermatology. AI has demonstrated impressive capabilities in image classification, object detection, and segmentation. Previous research has successfully trained machine learning models to classify Fitzpatrick skin type with accuracy rates between 81% and 96%.⁶ Other studies have explored AI-based classification of skin conditions such as oiliness and dryness.⁷ However, to date, no prior study has attempted to develop a machine learning model capable of simultaneously evaluating multiple diverse skin characteristics.

The SkinAnalysis Dataset and Machine Learning Model

A recent study aimed to bridge that gap by creating a novel dataset, SkinAnalysis, consisting of 3,662 images labeled with Fitzpatrick skin type, hyperpigmentation, redness, and wrinkle severity.⁸ These images were sourced from publicly available datasets, ensuring diversity in skin tones, ages, backgrounds, and lighting conditions. A double board-certified dermatologist annotated the dataset using standardized dermatologic scales, making it a robust foundation for machine learning model training.

To analyze the dataset, the study employed 3 established machine learning architectures: VGG-16, ResNet-50, and EfficientNet, all of which were pretrained on ImageNet. The final model was trained using a specialized loss function, SkinCELoss, which aligned with dermatologic scales and prevented inconsistent predictions. Data augmentation techniques such as random translations, rotations, and flips were applied to enhance the model's robustness.

Results and Implications

The study's best-performing model, an EfficientNet-V2M architecture, achieved a mean validation accuracy of 85.02% and an AUROC score of 0.8191. The test set results confirmed the model’s strong generalization capabilities, with a mean accuracy of 85.41% and an AUROC of 0.8306. The study also identified a trend where the model performed better at extreme values of each dermatologic scale but exhibited lower accuracy in middle-range classifications, likely due to increased clinical ambiguity.

These findings suggest that AI-driven dermatologic assessment models can provide dermatologist-level expertise in skin analysis, potentially enhancing the precision of laser therapy planning. By integrating such models into nonphysician-operated settings, medical spas and other aesthetic practices may improve their safety and treatment effectiveness.

Conclusion

This study represents a significant advancement in AI-driven dermatology by developing a machine learning model capable of assessing multiple skin characteristics simultaneously. The model’s high accuracy and generalizability suggest that AI has the potential to support nonphysician providers in making more precise skin assessments, thereby reducing complications associated with improper laser treatments. Researchers suggested future research should focus on expanding the dataset, refining AI-based treatment planning, and integrating AI-driven tools into clinical workflows to enhance patient safety and treatment efficacy. By leveraging AI in dermatologic assessments, the field can move toward more personalized, effective, and safer cosmetic procedures.

References

1. Butani A, Dudelzak J, Goldberg DJ. Recent advances in laser dermatology. J Cosmet Laser Ther. 2009;11(1):2-10. doi:10.1080/14764170802524411
2. Consumer survey on cosmetic dermatologic procedures. ASDS. 2021. Accessed March 27, 2025. https://www.asds.net/medical-professionals/practice-resources/consumer-survey-on-cosmetic-dermatologic-procedures.
3. Valiga A, Albornoz CA, Chitsazzadeh V, et al. Medical spa facilities and nonphysician operators in aesthetics. Clin Dermatol. 2022;40(3):239-243. doi:10.1016/j.clindermatol.2021.11.007
4. Occupational employment and wages. US Bureau of Labor Statistics, Occupational Employment Statistics, no. 2. Published 2019. Accessed March 27, 2025. https://www.bls.gov/oes/current/oes535011.htm#nat.
5. Rossi AM, Wilson B, Hibler BP, Drake LA. Nonphysician practice of cosmetic dermatology: A patient and physician perspective of outcomes and adverse events. Dermatol Surg. 2019;45(4):588-597. doi: 10.1097/DSS.0000000000001829
6. Chang CC, Hsing S, Chuang Y, et al. Robust skin type classification using convolutional neural networks. 2018 13th IEEE Conference on Industrial Electronics and Applications (ICIEA), Wuhan, China, 2018, pp. 2011-2014, doi: 10.1109/ICIEA.2018.8398040.
7. Saiwaeo S, Arwatchananukul S, Mungmai L, Preedalikit W, Aunsri N. Human skin type classification using image processing and deep learning approaches. Heliyon, Volume 9, Issue 11, 2023, e21176, ISSN 2405-8440. doi: 10.1016/j.heliyon.2023.e21176.
8. Draelos RL, Kesty CE, Kesty KR. Artificial intelligence predicts Fitzpatrick skin type, pigmentation, redness, and wrinkle severity from color photographs of the face. J Cosmet Dermatol. 2025;24(4):e70050. doi:10.1111/jocd.70050