Abstract

Idiopathic intracranial hypertension (IIH) requires longitudinal monitoring of disease activity and its retinal effects, while direct pressure assessment by lumbar puncture is invasive and impractical to repeat frequently. Optical coherence tomography (OCT) offers a non-invasive retinal adjunct, and hyperreflective foci (HRF) on OCT have emerged as a biomarker of clinical interest, making quantitative HRF readouts a candidate OCT-derived measure for investigation alongside pressure-based assessment. Accurate automated HRF segmentation, however, poses an asymmetric sensitivity–specificity dilemma beyond ordinary class imbalance: lesions occupy under 1% of a B-scan and are easily missed, while shifting toward sensitivity also admits speckle and vessel-like confounders under ambiguous annotations. We propose DiSS, a phased self-teaching scheme that, rather than competing the two goals within one objective, separates them in the training schedule on an unmodified backbone. A multi-timescale exponential moving average (EMA) teacher bank supplies progressively consolidated evidence; a first “cover” phase consolidates weak-lesion evidence via disjunctive (noisy-OR) fusion, and a second “suppress” phase applies soft-weighted residual modulation (SWRM) for evidence-guided spatial suppression and cyclic anchored snapshot self-teaching (CASST) for temporally anchored self-distillation to attenuate unsupported responses. Across two OCT datasets DiSS improves overlap, recall, and boundary metrics over the evaluated baselines; on an independent IIH cohort, DiSS-derived HRF readouts correlate with lumbar-puncture opening pressure more strongly than baseline-derived readouts (a directional, non-significant trend), and within-patient longitudinal changes in HRF burden track changes in neuroretinal layer thickness, while broader clinical validation is still required.

Links to Paper and Supplementary Materials

Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/OMIA_010.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=6rRi7ecXVY

BibTex

@InProceedings{WanZhi_APhased_MICCAISAT2026,
        author = { Wang, Zhi AND Wu, Donghan AND Hao, Huaying AND Cao, Le AND Wu, Bo AND Zhao, Yitian},
        title = { { A Phased Self-Teaching Framework for Hyperreflective Foci Segmentation in OCT } },
        booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
        year = {2026},
        publisher = {Springer Nature Switzerland},
        volume = {LNCS 17270},
        month = {pending},
        page = {pending}
}


back to top