Comparing Deep Learning Models for Lung Nodule Detection on Photon-Counting CT Images in a Non-Screening…

Abstract

BackgroundComprehensive evaluations of lung nodule detection methods remain limited, particularly regarding data from clinical routine, novel photon-counting CT (PCCT) technology, and the comparison of academic and commercial models.

Materials and MethodsThis retrospective study collected 1860 routine PCCT scans acquired at Hannover Medical School from 2021 to 2024. We compared four deep-learning-based lung nodule detection models, which were developed for conventional CT, namely the publicly available TotalSegmentator, nnDetection, and RadYOLO methods, as well as a commercial lung computer-aided detection (CAD) system. We evaluated the inter-model agreement in an unannotated cohort (n=1699), sensitivity in a smaller cohort of 25 manually annotated scans, and false positives as detections in 136 report-negative scans.

ResultsIn the unannotated cohort, the commercial CAD system detected 7071, TotalSegmentator 4264, nnDetection 5254, and RadYOLO 3122 nodules. Overall, the inter-model agreement was low, and the models produced substantially different candidate nodules. Approximately 1300 nodules were detected by all models, representing 18.7% - 42.8% of each models detections. 15.4% - 37.4% of a models detections did not match any other model. The models showed moderate sensitivity (between 0.71, 95%-CI [0.64, 0.77], and 0.83, 95%-CI [0.77, 0.88]), and TotalSegmentator and RadYOLO produced few false positives per scan (both 0.85, 95%-CI [0.70, 1.02]).

ConclusionThe lung nodules detected by four deep learning models showed high variation and a lack of consensus on clinical PCCT data. These findings highlight the importance of model benchmarking and local validation when deploying models in real-world clinical workflows.

Extracting smoking history from clinical notes for lung cancer screening decision support: comparing a structured-judgment model with general-purpose large language models
A. Wright, S. Liu, A. P. Wright
medRxiv24 Sep 2026PreprintAll rights reservedFull textEpidemiology & public healthmedRxiv
Empirical Antibiotic Initiation for Suspected Early-Onset Neonatal Infection in the Setting of Maternal GBS Screening and Intrapartum Antibiotic Prophylaxis: A Single-Center Retrospective Cohort Study
Anna Zmelonek-Znamirowska, Grzegorz Świercz, Jakub Mlodawski +2
Medicina (Kaunas, Lithuania)1 Oct 2026Peer-reviewedCC BYFull textGeneral medicineEurope PMC
Identification of key clinical features and development of a machine learning model for screening-based detection of diabetic retinopathy
Li Huilin, Hao Shaofeng, Li Na +4
Scientific Reports1 Jul 2026Peer-reviewedCC BYFull textMedical imaging & radiologyDOAJ
Mitochondrial DNA depletion syndrome 12B: role of photon-counting computed tomography in detection of cardiac involvement—a case report
European heart journal. Case reports19 Sep 2026Peer-reviewedCC BYFull textCardiovascular medicineEurope PMC
Evaluation of iodine detection and quantification limits in photon-counting CT—a phantom study
Rebecca Titternes, Gavin Poludniowski, Torkel Bo Brismar +2
European radiology experimental12 Oct 2026Peer-reviewedCC BYFull textMedical imaging & radiologyEurope PMC

Cite

BibTeX RIS

BibTeX

@misc{thieme2026comparing,
  title = {{Comparing Deep Learning Models for Lung Nodule Detection on Photon-Counting CT Images in a Non-Screening Setting}},
  author = {L. Thieme and E. Petersen and K. Geissler and L. Müller-Groh and S. Oeltze-Jafra and H.-o. Shin and S. C. Scharm and F. Wacker and Z. Ahmadi and A. Schenk},
  howpublished = {medRxiv},
  year = {2026},
  month = sep,
  doi = {10.64898/2026.09.28.26364159},
  url = {https://www.medrxiv.org/content/10.64898/2026.09.28.26364159v1},
  note = {CC BY},
}

RIS

TY  - UNPB
TI  - Comparing Deep Learning Models for Lung Nodule Detection on Photon-Counting CT Images in a Non-Screening Setting
AU  - L. Thieme
AU  - E. Petersen
AU  - K. Geissler
AU  - L. Müller-Groh
AU  - S. Oeltze-Jafra
AU  - H.-o. Shin
AU  - S. C. Scharm
AU  - F. Wacker
AU  - Z. Ahmadi
AU  - A. Schenk
PY  - 2026
DA  - 2026/09/28
PB  - medRxiv
DO  - 10.64898/2026.09.28.26364159
UR  - https://www.medrxiv.org/content/10.64898/2026.09.28.26364159v1
AB  - BackgroundComprehensive evaluations of lung nodule detection methods remain limited, particularly regarding data from clinical routine, novel photon-counting CT (PCCT) technology, and the comparison of academic and commercial models.  Materials and MethodsThis retrospective study collected 1860 routine PCCT scans acquired at Hannover Medical School from 2021 to 2024. We compared four deep-learning-based lung nodule detection models, which were developed for conventional CT, namely the publicly available TotalSegmentator, nnDetection, and RadYOLO methods, as well as a commercial lung computer-aided detection (CAD) system. We evaluated the inter-model agreement in an unannotated cohort (n=1699), sensitivity in a smaller cohort of 25 manually annotated scans, and false positives as detections in 136 report-negative scans.  ResultsIn the unannotated cohort, the commercial CAD system detected 7071, TotalSegmentator 4264, nnDetection 5254, and RadYOLO 3122 nodules. Overall, the inter-model agreement was low, and the models produced substantially different candidate nodules. Approximately 1300 nodules were detected by all models, representing 18.7% - 42.8% of each models detections. 15.4% - 37.4% of a models detections did not match any other model. The models showed moderate sensitivity (between 0.71, 95%-CI [0.64, 0.77], and 0.83, 95%-CI [0.77, 0.88]), and TotalSegmentator and RadYOLO produced few false positives per scan (both 0.85, 95%-CI [0.70, 1.02]).  ConclusionThe lung nodules detected by four deep learning models showed high variation and a lack of consensus on clinical PCCT data. These findings highlight the importance of model benchmarking and local validation when deploying models in real-world clinical workflows.
ER  -

Abstract

Related

Cite