Abstract
Pulmonary infection is a major cause of morbidity and mortality in immunocompromised patients. The causative organisms include fungi, bacteria and mycobacteria requiring different treatments, yet their nodular appearances on CT often overlap. We study deep-learning classification of four pulmonary infections in a private cohort of 196 CT scans: Aspergillosis (n = 123), Tuberculosis (n = 41), Nocardiosis (n = 16) and Mucormycosis (n = 16). Comparing ImageNet-pretrained CNN and Transformer backbones and fine-tuning regimes, we find similar performance across architectures, with Swin-T selected as the best-performing backbone (macro AUROC 0.71). Although it predicts Aspergillosis for most scans and never predicts either rare class, out-of-fold predictions retain discriminative signal: Mucormycosis achieves the highest per-class AUROC (0.76, p = 0.0002) despite zero recall. A two-stage analysis localises the main difficulty to separating the rare infections from Aspergillosis: with correct routing, the best classifier achieves recalls of 0.62 for Nocardiosis and 0.38 for Mucormycosis, while fixed Swin-T still misroutes 38% of non-Aspergillosis scans. Thus, rare infections are not simply unrecognisable from CT; their discriminative signal is not reliably converted into multiclass predictions.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/TIA_028.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{BisSup_Deep_MICCAISAT2026,
author = { Biswas, Suparna AND Ebou, El Haj Samitt AND Mahiou, Yanni AND Fournier, Laure AND Duron, Loïc AND Martin, Garance AND de Margerie-Mellon, Constance},
title = { { Deep Learning Classification for Rare Pulmonary Infection Differentiation in Immunocompromised Patients: A Data-Limited Study } },
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}
}