Deep-learning-based simulation study of aesthetic outcomes in microplasma radiofrequency ablation for pathological scars
LI Chengfei
LI Shiyi
FU Qiang
ZHOU Guiwen
WU Qian
MENG Fanting
XU Xiao
CHEN Minliang
Abstract：Objective To develop a GAN-based system to simulate post-treatment appearance after micro-plasma radiofre-quency(MPR)therapy for pathological scars,enabling visual outcome prediction from a single pre-treatment clinical photograph to support counseling and individualized decision-making.Methods Standardized paired photographs were retrospectively collected from patients receiving MPR monotherapy between 2020 and 2024,with follow-up images acquired at 6 months.A total of 100 cases were included and split into a training set(n=80)and a test set(n=20).A paired,conditional GAN was built with a ResNet-based encoder decoder generator(9 residual blocks)and a PatchGAN discriminator,optimized using adversarial loss plus a weighted L1 loss(λ=100).Agreement between generated and ground-truth follow-up images was evaluated using MSE,PSNR,and SSIM,along with representative qualitative examples.Results The model reproduced major post-MPR visual trends,including preserved scar morphology,smoother margins,and reduced ery-thema.Test-set performance was:MSE(0.053±0.024)95%CI(0.042,0.064),PSNR(37.069±3.215)dB 95%CI(35.564,38.574),and SSIM(0.736±0.036)95%CI(0.719,0.753).Conclusion The proposed system is feasible for simulating post-MPR scar appearance and may serve as a visual aid for pre-treatment counseling and decision support.Further multicenter validation and integration of clinical pa-rameters and multimodal data are warranted.
Keywords：Pathological scarMicro-plasma radiofrequencyGenerative adversarial networkImage-to-image translationOut-come simulation
Publication Date：2026-02-15
Online Publishing Date：2026-03-31(First online date of this platform, not the publication date of the document)
Pages：8( 100-106,后插6 )