Abstract
Spirometry is the clinical standard for pulmonary-function assessment, but its results depend on patient effort, operator expertise, and test availability. Chest CT provides effort-independent structural information, yet direct volumetric learning usually requires large annotated cohorts. We present GPT-DBR, a small-cohort experimental framework that applies decoding-based language-model regression to standardized CT-derived biomarkers after quantitative CT and radiomic feature extraction. In 164 paired CT–spirometry cases, GPT-DBR($\mathrm{qCT}$) achieved the lowest $\mathrm{FEV}_1/\mathrm{FVC}\%$ error ($\mathrm{MAE}=7.81$) and competitive $\mathrm{FEV}_1\%$ predicted error ($\mathrm{MAE}=15.28$). In the held-out split, the secondary classification AUC for $\mathrm{FEV}_1/\mathrm{FVC}<0.70$ increased from $0.469$ for GPT-base($\mathrm{qCT}$) to $0.768$ for GPT-DBR($\mathrm{qCT}$). These proof-of-concept findings support further investigation of language-model regression heads for structured CT biomarkers in data-scarce cohorts, while external validation remains necessary before clinical translation.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/EMERGE_027.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=VJcM5fvlrt
BibTex
@InProceedings{KimYou_GPTDBR_MICCAISAT2026,
author = { Kim, YoungSeok AND Park, Kwangsuk AND Jo, WoongJae AND Hong, Yerin AND Kim, Yun-Hyeon},
title = { { GPT-DBR: Decoding-Based Language-Model Regression for CT-Derived Lung-Function Estimation } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17266},
month = {pending},
page = {pending}
}