Abstract
Ultrasound-based ovarian tumor classification using deep learning is often affected by diagnostic bias caused by non-biological measurement artifacts and severe class imbalance, which introduce shortcut correlations and reduce sensitivity to rare but clinically important tumor types. These limitations hinder the robustness and clinical reliability of automated diagnostic systems. In this study, we present a unified framework that jointly mitigates visual and distributional biases for robust ovarian ultrasound classification. The proposed approach integrates segmentation-guided inpainting to automatically detect and remove measurement overlays while preserving tumor morphology and intrinsic ultrasound texture, together with imbalance-aware learning that combines data rebalancing and a hybrid optimization objective to improve minority-class discrimination.We evaluate the framework on the MMOTU-2D dataset containing multiple ovarian tumor subtypes with long-tailed distribution. Experimental results demonstrate consistent improvements across multiple convolutional backbones, with the full pipeline achieving 80.38\% Top-1 accuracy using EfficientNet-B2 and outperforming baseline models with lower computational complexity. Ablation studies confirm that artifact removal and imbalance-aware optimization address complementary failure modes, leading to improved robustness and generalization. This work highlights the importance of jointly correcting visual and distribution biases to enable reliable and clinically applicable ultrasound-based tumor classification.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/MIRASOL_019.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=7WYDGcscyU
BibTex
@InProceedings{KhaAla_SegmentationGuided_MICCAISAT2026,
author = { Khaled, Alaa AND Hatem, Malak AND Eissa, Malak AND Abozeid, Sara AND Selim, Sahar},
title = { { Segmentation-Guided Inpainting and Imbalance-Aware Learning for Robust Ovarian Tumor Classification in Ultrasound } },
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}
}