Abstract
Automated report generation from non-contrast head CT scans is essential for reducing the workload on radiologists, but challenges remain in processing the spatial context of 3D data and performing many-to-many alignment between images and text. In this paper, we propose a highly efficient multimodal model based on the MICCAI 2026 HEADLINE Challenge. The method consists of a pre-trained HLIP visual encoder, a two-layer MLP projector, and a 4-bit quantized Qwen3.5-4B model. It achieves stable spatial alignment while minimizing computational resources with LoRA and a two-stage training strategy. Experiments confirm a BERTScore of 0.8632 on the validation set and an overall score of 0.2403 in the official validation phase.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/HEADLINE_004.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=k47C5k0hMw
BibTex
@InProceedings{KonSat_EHead_MICCAISAT2026,
author = { Kondo, Satoshi AND Kasai, Satoshi},
title = { { E-Head: Efficient Head CT Report Generation via Multi-Resolution Series Integration and Two-Stage Parameter-Efficient Fine-Tuning } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17263},
month = {pending},
page = {pending}
}