CXRL

Scalp Diagnostic System with Label-Free Segmentation and Training-Free Image Translation

MICCAI 2025
Youngmin Kim1*, Saejin Kim1*, Hoyeon Moon1, Youngjae Yu2+, Junhyug Noh3+
1 Yonsei University  2 Seoul National University  3 Ewha Womans University
Paper arXiv Code
Overview of ScalpVision

We introduce ScalpVision,
Scalp Diagnostic System with Label-Free Segmentation and Training-Free Image Translation.

Abstract

Dermatoses of the scalp affect millions of people around the world, underscoring the urgent need for early diagnosis and management of the disease. However, the development of a comprehensive AI-based diagnosis system encompassing these conditions remains an underexplored domain due to the challenges associated with data imbalance and the costly nature of labeling. To address these issues, we propose ScalpVision, an AI-driven system for the holistic diagnosis of scalp diseases. In ScalpVision, effective hair segmentation is achieved using pseudo image-label pairs and an innovative prompting method in the absence of traditional hair masking labels. Additionally, ScalpVision introduces DiffuseIT‑M, a generative model adopted for dataset augmentation while maintaining hair information, facilitating improved predictions of scalp disease severity. Our experimental results affirm ScalpVision's efficiency in diagnosing a variety of scalp conditions, showcasing its potential as a valuable tool in dermatological care.

Method


Pipeline

Main figure

A high-level view of the full system. We first obtain a reliable hair mask without manual labels, then use it to steer training-free diffusion for data augmentation, and finally train a scalp-condition classifier on real+augmented images.

  • Label-Free Hair Masking (M̂ → MAP → M): Train a naive segmenter on pseudo image–label pairs to get a coarse mask M̂, generate automatic positive/negative point prompts from M̂ for SAM to produce MAP, then AND-ensemble (M = M̂ ∧ MAP) and clean small components to form a robust hair mask.
  • Mask-Guided Translation (DiffuseIT-M): Perform training-free diffusion editing that applies style/semantic changes only to scalp pixels (1−M) while freezing hair (M). A composite objective (style, content, semantic, mask-preservation, range) guides the reverse denoising so hairlines and fiber texture remain intact.
  • Scalp Condition Diagnosis: Use real images + DiffuseIT-M augmentations to address class imbalance and train a classifier that detects dandruff, excess sebum, erythema and their severity (good/mild/moderate/severe)—improving robustness on under-represented conditions.

Qualitative Results


Hair segmentation comparison between previous models and our method

segmentation comparision

Comparison of various segmentation methods on hair. "GT" represents the mask images for which we have manually annotated the pixel segmentation.


Scalp augmentation comparison between previous models and our method

translation results

Microsocpic scalp image translation results with different generative models, where the goal is preserving source hairlines while changing the scalp.


Additional Qualitative results

mask guidance

Image translation results using various mask guidance.

Qualitative Results


Hair Segmentation and Translation Results

segmentation-table translation-table

Performance comparison of hair segmentation and scalp image translation on the test set.


Classification Results

ablation

Performance of scalp condition classification with various augmentation methods, denoted after "+" symbol, on the test set. The second column displays the overall macro-F1 score, while the columns from the third onward show the F1 scores for each severity level of the three diseases

BibTeX

@InProceedings{kim2025scalpvision,
      author    = {Kim, Youngmin and Kim, Saejin and Moon, Hoyeon and Yu, Youngjae and Noh, Junhyug},
      title     = {Scalp Diagnostic System with Label-Free Segmentation and Training-Free Image Translation},
      booktitle = {International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI)},
      month     = {Sep},
      year      = {2025}
}