Machine Learning Based Investigation of Relation Between Patient and Glabellar Wrinkle Characteristics

Year 2025, Volume: 22 Issue: 2, 381 - 387, 27.06.2025

Çağlar Cengizler , Ayşe Gül Kabakcı , Dilek Eren , Dursun Murat Bozkır , Eda Esra Esen , Yaşam Türközer , Memduha Gülhal Bozkır

https://doi.org/10.35440/hutfd.1689703

Abstract

Background: The skin around the glabella area is the region most affected by the underlying muscle activity during facial expressions and wrinkling. Although glabellar lines or wrinkles are commonly considered a cosmetic concern in the facial region, objective clinical assessment of this condition remains important. This study investi-gated the relationship between various patient characteristics, potentially reflecting lifestyle and physiological factors, and glabellar wrinkle patterns, which were categorized according to distinct muscle contraction types described in the literature.
Materials and Methods: Data were collected from a total of 870 patients. One of the study's primary objectives was to explore the potential of using multiple patient-related variables in a machine learning-based prediction system while identifying the most influential characteristics associated with glabellar wrinkles. Multiple super-vised machine learning algorithms were employed to uncover potentially nonlinear associations and identify the most informative predictors, including Naïve Bayes, Random Forest, Logistic Regression, and K-Nearest Neigh-bors. Each model was tested across all possible feature subsets to evaluate their predictive performance.
Results: The study aimed to demonstrate that a machine learning-based approach could be functional for the early prediction of glabellar wrinkles. The analysis revealed significant associations between patient characteris-tics, such as age, gender, education, marital status, and occupation, and distinct glabellar wrinkle patterns. Fur-thermore, the machine learning models demonstrated that these characteristics could be used to predict wrin-kle types with considerable performance. In the best scenario, separating the U-type pattern from others, an F1 score of 0.71 was achieved using K-Nearest Neighbors, supporting the potential for early identification and personalized intervention planning.
Conclusions: The study confirmed that specific patient characteristics are strongly related to the formation of different glabellar wrinkle types. In addition, machine learning-based predictive systems showed promising performance, indicating their potential use in supporting clinicians with personalized cosmetic assessments and early intervention strategies.

Keywords

Artificial Intelligence , Skin Aging , Machine Learning , Demography

Ethical Statement

This study was performed in line with the principles of the Declaration of Helsinki. Ethical approval was obtained from the Non-Interventional Clinical Research Ethics Committee of the Faculty of Medicine at Çukurova University on June 2, 2023 (approval number 134/44).

Supporting Institution

Çukurova University Scientific Research Projects Unit

Project Number

TSA-2024-16787

Thanks

We would like to thank to Çukurova University Scientific Research Projects Unit (project number TSA-2024-16787).

Hasta Özellikleri ile Glabellar Kırışıklık İlişkisinin Makine Öğrenimi Tabanlı İncelenmesi

Abstract

Amaç: Glabella bölgesindeki deri, yüz ifadesi ve kırışıklık oluşumu sırasında alttaki kas aktivitesinden en çok etkilenen alanlardan birisidir. Glabellar çizgiler veya kırışıklıklar genellikle kozmetik kaygı nedeni olabilmekle birlikte, bu çizgi ve kırışıklıkların klinik açıdan nesnel değerlendirilmesi önem taşımaktadır. Bu çalışmada, hastala-rın yaşam tarzı ve fizyolojilerini yansıtabilecek çeşitli özellikleri ile literatüre göre sınıflandırılmış glabellar kırışıklık karakterleri arasındaki ilişki araştırılmıştır.
Materyal ve Metod: Bu araştırmada toplam 870 hastadan veri toplanmıştır. Çalışmada öncelikli olarak hastaları karakterize eden birden fazla değişkenin makine öğrenmesi tabanlı bir tahmin-teşhis sisteminde kullanılabilirliği-nin araştırılması hedeflenmiş, bu doğrultuda glabellar kırışıklıklarla ilişkili olabilecek en etkili özelliklerin belirlen-mesi amaçlanmıştır. Naïve Bayes, Rastgele Orman, Lojistik Regresyon ve K-En Yakın Komşular gibi farklı kılavuzlu makine öğrenmesi algoritmaları kullanılarak, doğrusal olmayan potansiyel ilişkilerin ortaya çıkarılması hedeflen-miş, en ilintili değişkenler belirlenmiştir. Her modelin tahmin performansı, tüm olası özellik alt kümeleri üzerinde test edilmiştir.
Bulgular: Çalışmada, glabellar kırışıklık karakterinin önceden tahmini için makine öğrenimi temelli bir yaklaşımın işlevsel olabileceğinin gösterilmesi amaçlanmıştır. Analiz sonuçları; yaş, cinsiyet, eğitim, medeni durum ve meslek gibi hasta özellikleri ile glabellar kırışıklık karakteri arasında önemli ilişkiler olduğunu göstermektedir. Ayrıca çalışmada, bu özelliklerle kırışıklık tiplerinin öngörülebilirliği araştırılmış, U tipi desen için K-En Yakın Komşu algoritmasıyla 0.71 F1 skoru elde edilmiştir. Bu sonuç, erken teşhis ve kişiselleştirilmiş planlama açısından yönte-min potansiyelini ortaya koymaktadır.
Sonuç: Çalışma, belirli hasta özelliklerinin farklı glabellar kırışıklık türlerinin oluşumuyla önemli derecede ilintili olduğunu doğrulamıştır. Ayrıca makine öğrenimi tabanlı potansiyel bir teşhis ve kestirim sisteminin bu özellikler ile yüksek performanslı sınıflama yapması otomatik sınıflayıcıların kişiselleştirilmiş kozmetik değerlendirme ve erken müdahale için destek potansiyeline işaret etmektedir.

Keywords

Makine Öğrenmesi , Yapay Zekâ , Cilt Yaşlanması , Demografi

Project Number

TSA-2024-16787

References

• 1. Kim HS, Kim C, Cho H, Hwang JY, Kim YS. A study on glabellar wrinkle patterns in Koreans. J Eur Acad Dermatol Venereol. 2014;28(10):1332-1339.
• 2. Lewis MB, Bowler PJ. Botulinum toxin cosmetic therapy correlates with a more positive mood. J Cosmet Dermatol. 2009;8(1):24-26.
• 3. de Almeida ART, da Costa Marques ER, Kadunc BV. Glabellar wrin-kles: a pilot study of contraction patterns. Surg Cosmet Dermatol. 2010;2(1):23-28.
• 4. Carruthers J, Fagien S, Matarasso SL, Botox Consensus Group. Con-sensus recommendations on the use of botulinum toxin type A in facial aesthetics. Plast Reconstr Surg. 2004;114(6):1S-22S.
• 5. Draelos ZD. The shrinking world: skin considerations in a global community. J Cosmet Dermatol. 2006;5(1):1-2.
• 6. Rexbye H, Petersen I, Johansens M, Klitkou L, Jeune B, Christensen K. Influence of environmental factors on facial ageing. Age Ageing. 2006;35(2):110-115.
• 7. Shah SAA, Bennamoun M, Molton MK. Machine learning approaches for prediction of facial rejuvenation using real and synthetic data. IEEE Access. 2019;7:23779-23787.
• 8. Yap MH, Batool N, Ng CC, Rogers M, Walker K. A survey on facial wrinkles detection and inpainting: datasets, methods, and challeng-es. IEEE Trans Emerg Top Comput Intell. 2021;5(4):505-519.
• 9. Kim S, Yoon H, Lee J, Yoo S. Semi-automatic labeling and training strategy for deep learning-based facial wrinkle detection. In: Pro-ceedings of IEEE CBMS. IEEE; 2022. p. 383-388.
• 10. de Lima AS, Gubert M, Lyra T, Sardagna CF. Muscle contraction patterns and their associations in the upper third of the face: a clini-cal and epidemiological study. Surg Cosmet Dermatol. 2022;14:1-10.
• 11. Soria D, Garibaldi JM, Ambrogi F, Biganzoli EM, Ellis IO. A ‘non-parametric’ version of the naive Bayes classifier. Knowl Based Syst. 2011;24(6):775-784.
• 12. Kononenko I. Semi-naive Bayesian classifier. In: Kodratoff Y, editor. Machine Learning—EWSL-91: European Working Session on Learn-ing; 1991 Mar 6-8; Porto, Portugal. Berlin Heidelberg: Springer; 1991. p. 206-219.
• 13. Azar AT, Elshazly HI, Hassanien AE, Elkorany AM. A random forest classifier for lymph diseases. Comput Methods Programs Biomed. 2014;113(2):465-473.
• 14. Pal M. Random forest classifier for remote sensing classification. Int J Remote Sens. 2005;26(1):217-222.
• 15. De Menezes FS, Liska GR, Cirillo MA, Vivanco MJ. Data classification with binary response through the Boosting algorithm and logistic re-gression. Expert Syst Appl. 2017;69:62-73.
• 16. Hajmeer M, Basheer I. Comparison of logistic regression and neural network-based classifiers for bacterial growth. Food Microbiol. 2003;20(1):43-55.
• 17. Halder RK, Uddin MN, Uddin MA, Aryal S, Khraisat A. Enhancing K-nearest neighbor algorithm: a comprehensive review and perfor-mance analysis of modifications. J Big Data. 2024;11(1):113.
• 18. Laaksonen J, Oja E. Classification with learning k-nearest neighbors. In: Proceedings of International Conference on Neural Networks (ICNN'96). IEEE; 1996. p. 1480-1483.
• 19. Luebberding S, Krueger N, Kerscher M. Quantification of age-related facial wrinkles in men and women using a three-dimensional fringe projection method and validated assessment scales. Dermatol Surg. 2014;40(1):22-32.
• 20. Vierkötter A, Krutmann J. Environmental influences on skin aging and ethnic-specific manifestations. Dermato-endocrinol. 2012;4(3):227-231.