Abstract
Automated localisation of hepatocellular carcinoma (HCC) on contrast-enhanced CT scans is difficult when lesions are small, weakly enhancing, or obscured by heterogeneous liver parenchyma. Although 2D detectors provide efficient tumour localisation, analysing each axial slice independently can produce inconsistent predictions across a CT volume. We address this limitation with L-Spot, a slice-aware consensus framework that refines detector outputs by combining complementary YOLO predictions with spatial agreement across neighbouring slices. The framework uses two trained detectors: YOLOv8m, evaluated under both standard and augmented inference, and an independently trained YOLOv10m model. Predicted lesions are retained only when multiple prediction sources agree at anatomically consistent locations. We evaluated L-Spot on arterial-phase CT scans from 94 HCC patients and found that it improved the precision–recall balance over the strongest single-detector baseline by suppressing isolated false positives while preserving true positive tumour detections. L-spot achieved an F1-score of 0.733 (precision: 0.820, recall: 0.664), representing an 11.8% improvement over the baseline YOLOv8m. These results demonstrate that detector consensus and inter-slice continuity can improve the robustness of HCC localisation without altering the underlying detection architectures.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/CaPTion_032.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=~VIVEK_KUMAR_SINGH6
BibTex
@InProceedings{KamJul_LSpot_MICCAISAT2026,
author = { Kampel, Julia Rebecca AND Singh, Tripti AND Akram, Farhan AND Jegiraj, Anne Christina Melissa AND Bommaraveni, Shriya AND Morris, Byron John AND Tiwari, Akshat AND Deshpande, Sheetal Santosh AND Shahabuddin, Khawaja AND Puig, Domenec AND Singh, Vivek Kumar},
title = { { L-Spot: Slice-Aware Consensus for Hepatocellular Carcinoma Localisation in Contrast-Enhanced CT } },
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}
}