Abstract
Accurate segmentation of endometriosis lesions in laparoscopic images remains challenging due to subtle boundaries, heterogeneous appearances and locations, and limited annotated data. While SAM-based foundation models enable flexible promptable segmentation, their value for fully automatic AI-assisted endometriosis surgery remains unclear. We present a cross-complementary dataset benchmark of supervised, SAM-based promptable, and hybrid automatic approaches for laparoscopic endometriosis lesion segmentation. Experiments were conducted on ENID, dominated by implant-like lesions, and GLENDA-clean, containing more heterogeneous lesion appearances and anatomical locations. We evaluated SAM2, MedSAM, SAM-Med2D, and SurgiSAM2 without fine-tuning under four different controlled prompt settings. Box prompts were most reliable, increasing mean Dice over point prompts by 0.179 on ENID and 0.410 on GLENDA-clean, while additional positive or negative points did not significantly improve performance. SAM2 achieved the highest Dice on ENID, whereas SurgiSAM2 performed best on GLENDA-clean. We then trained DeepLabV3+, SegFormer, YOLO11s-seg, U-Net++ and nnU-Net v2, as fully automatic baselines, with SegFormer showing the strongest overall cross-dataset performance. Finally, as a hybrid experiment, we evaluated SegFormer-to-SurgiSAM2 automatic refinement using predicted lesion candidates as prompts. Fallback acceptance reduced degradation compared with direct refinement, but no hybrid variant outperformed SegFormer alone. Our code and benchmark is available at: https://github.com/JArjomandi/Auto_Endometriosis_Segmentation
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/CAPI_WOMEN_018.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=eTCvl34Ttr
BibTex
@InProceedings{ArjJas_Evaluating_MICCAISAT2026,
author = { Arjomandi, Jasmin AND Kunz, Christian AND Kang, Xingjian AND Breininger, Katharina AND Mathis-Ullrich, Franziska},
title = { { Evaluating the Prompt-to-Automation Gap: A Comprehensive Benchmark for Laparoscopic Endometriosis Segmentation } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17256},
month = {pending},
page = {pending}
}