Master Thesis

Status: open
Supervisor: Florian Kleber

The aim of this master’s thesis is to analyze trichoscopic scalp images in collaboration with a dermatology center. The focus is on

• Developing and evaluating algorithms for segmenting and counting hair follicles
• Automated measurement of basic parameters: number of follicles, hair shaft diameter, interfollicular density, follicle quality (single- or multi-terminal), vellus hair
• Robustness evaluation under real-world imaging conditions (without shaving, varying lighting)
• Further stages of the model are intended to enable the automated detection of inflammatory hair follicle diseases

To evaluate computer vision methods for hair image analysis, a compact benchmark can combine three complementary model classes. First, an object detection approach such as YOLOv8 can be used to localize and count hair follicles, providing a direct estimate of follicle density and spatial distribution based on bounding box predictions. Second, instance segmentation models such as Mask R-CNN or Mask2Former can be applied to delineate individual follicles or hair shafts at pixel level, enabling more precise morphometric measurements, including local thickness estimation and shape descriptors, while also addressing overlapping structures. Finally, a self-supervised foundation model baseline such as DINOv2 can be used as a feature extractor without task-specific supervision to provide a strong baseline.

The research consists of

• Literature Review – getting to know the methods
• Implementation & Evaluation
  • Evaluate state-of-the-art methods on the provided datasets
  • Develop and apply your processing pipeline
  • Comparison and thorough evaluation
• Written Thesis and final presentation
• Summarize your work in a publication (optionally).

References

Caro RDC, Orlova V, Meo ND, Zalaudek I. Analysis of trichoscopic images using deep neural networks for the diagnosis and activity assessment of alopecia areata – a retrospective study. J Dtsch Dermatol Ges. 2026 Jan;24(1):44-55. doi: 10.1111/ddg.15847. Epub 2025 Sep 30. PMID: 41025749; PMCID: PMC12800882.

Nguyen HV, Byeon H. YOLOv7-based automated detection platform for scalp lesions. Digit Health. 2024 Sep 2;10:20552076241279185. doi: 10.1177/20552076241279185. PMID: 39262419; PMCID: PMC11387947.

 Helpful experience

• Python
• Good understanding of deep learning
• Machine Learning frameworks (preferably PyTorch)
• Interest in deep learning