Abstract

Resource-efficient clinical prediction from medical images is particularly challenging when image quality is degraded by artifacts, and large annotated datasets or high-performance computing infrastructure are unavailable. In dermoscopic skin lesion analysis, hair occlusions can obscure diagnostically relevant structures and negatively impact automated lesion classification. We propose a single-image inpainting approach for removing hair artifacts from skin lesion photographs. This method trains an image-specific generative adversarial model using only non-occluded regions of the target image, thereby avoiding the need for dataset-scale training or task-specific pretrained restoration models. The resulting generator reconstructs occluded regions in a manner consistent with the local appearance of the lesion and surrounding skin. We evaluate the approach using both image restoration metrics and downstream lesion classification performance, demonstrating that dataset-free inpainting can serve as a lightweight preprocessing step for resource-efficient clinical prediction. Our results indicate that single-image generative restoration can enhance the accessibility of dermatological image analysis pipelines in settings with limited data and computational resources, addressing practical constraints and supporting clinical relevance.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MIRASOL_082.pdf

SharedIt Link: Not yet available

SpringerLink (DOI): Not yet available

Supplementary Material: Not Submitted

Link to Open Review

Open Review Page: https://openreview.net/forum?id=noYtOg09ar

BibTex

@InProceedings{KucDar_DatasetFree_MICCAISAT2026,
        author = { Kucharski, Dariusz AND Jaworek-Korjakowska, Joanna},
        title = { { Dataset-Free Dermoscopic Hair Removal } },
        booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
        year = {2026},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 17265},
        month = {pending},
        page = {pending}
}


back to top