Abstract
Longitudinal dermoscopic images are valuable for studying how melanocytic lesions change over time, but they are difficult to collect at scale. Suspicious lesions are often removed once concerning change is observed, and benign lesions under surveillance may show only subtle visible progression. We propose an ABCD-guided framework for controlled synthetic editing of dermoscopic lesions. Given a source image, a lesion bounding box or mask, and a text prompt describing asymmetry, border irregularity, color variegation, and diameter, the model generates an edited lesion appearance while preserving source-lesion information and local skin context. The framework is trained in two phases. First, an I-JEPA encoder is adapted to dermoscopic images using self-supervised learning and then frozen to provide lesion-state representations. Second, a trainable latent editor and diffusion decoder use these representations for prompt-conditioned lesion generation. We evaluate generated images using three complementary analyses: embedding-based identity preservation, trajectory behavior across controlled edit steps, and human assessment of perceptual realism, visible artifacts, clinical plausibility, and lesion identity preservation. Generated outputs are intended as controlled synthetic simulations of requested morphological changes, not as verified predictions of true biological lesion progression.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/CaPTion_030.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/profile?id=~Quang_Do_Duy1
BibTex
@InProceedings{DoQua_IdentityPreserving_MICCAISAT2026,
author = { Do, Quang AND Dang, Khoi Tran AND Nguyen, Hang},
title = { { Identity-Preserving Synthetic Dermoscopic Lesion Generation via ABCD Prompting } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17278},
month = {pending},
page = {pending}
}