Abstract
Interstitial lung diseases comprise a heterogeneous group of pulmonary disorders, among which idiopathic pulmonary fibrosis is of particular clinical importance due to its poor prognosis and its association with the usual interstitial pneumonia (UIP) pattern. Although high-resolution computed tomography (HRCT) plays a central role in the identification of UIP, visual interpretation remains challenging and subject to inter-observer variability. This work presents a systematic evaluation of several deep learning configurations for automated UIP detection from HRCT scans. The investigated approaches explore different input representations, feature extractors and aggregation strategies. Experiments were conducted on a private dataset comprising 663 HRCT examinations from 262 patients using a five-fold cross-validation protocol. The best-performing configuration, combining a 2D ResNet encoder with a global multiple instance learning strategy, achieved an AUC of 0.943. These findings highlight the potential of slice-level representations and learned aggregation mechanisms for robust and computationally efficient UIP detection from HRCT scans.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/TIA_025.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: Not Available
BibTex
@InProceedings{MelTân_Deep_MICCAISAT2026,
author = { Melo, Tânia AND Fernandes, João AND Sousa, Célia AND Mendonça, Ana Maria AND Pedrosa, João},
title = { { Deep Learning for UIP Detection in HRCT Scans: A Comparative Evaluation of Input Representations, Feature Extractors, and Aggregation Strategies } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17269},
month = {pending},
page = {pending}
}