CaltechAUTHORS
Published June 2025 | Version Published
Journal Article Open

Deep learning-based detection and segmentation of osseous metastatic prostate cancer lesions on computed tomography

Creators

  • 1. ROR icon University of Southern California
  • 2. ROR icon California Institute of Technology
  • 3. ROR icon Children's Hospital of Los Angeles

Abstract

Introduction

Prostate adenocarcinoma frequently metastasizes to bone and is detected via computed tomography (CT) scans. Accurate detection and segmentation of these lesions are critical for diagnosis, prognosis, and monitoring. This study aims to automate lesion detection and segmentation using deep learning models.

Methods and Materials

We evaluated deep learning models for lesion detection (EfficientNet, ResNet34, DenseNet) and segmentation (nnUNetv2, UNet, ResUNet, ResAttUNet). Performance metrics included F1 score, precision, recall, Area Under the Curve (AUC), and Dice Similarity Coefficient (DSC). Pairwise t-tests compared segmentation accuracy. Radiomic analyses compared lesions segmented by deep learning to manual segmentations.

Results

EfficientNet achieved the highest detection performance, with an F1 score of 0.82, precision of 0.88, recall of 0.79, and AUC of 0.71. Among segmentation models, nnUNetv2 performed best, achieving a DSC of 0.74, precision of 0.73, and recall of 0.83. Pairwise t-tests showed that nnUNetv2 outperformed other models in segmentation accuracy (p < 0.01). Clinically, nnUNetv2 also demonstrated superior specificity for lesion detection (0.90) compared to the other models. All models performed similarly in distinguishing diffuse and focal lesions, predicting weight-bearing lesions, and identifying lesion locations, although nnUNetv2 had higher specificity. Sensitivity was highest for rib lesions and lowest for spine lesions across all models.

Conclusions

EfficientNet and nnUNetv2 were the top-performing models for detection and segmentation, respectively. Radiomic features derived from deep learning-based segmentations were comparable to manual segmentations, supporting clinical applicability. Further analysis of lesion detection and spatial distribution underscores the models' potential for improving diagnostic workflows and patient outcomes.

Copyright and License (English)

© 2025 The Author(s). Published by Elsevier B.V. This article is available under the Creative Commons CC-BY-NC-ND license and permits non-commercial use of the work as published, without adaptation or alteration provided the work is fully attributed.

Funding (English)

This work was supported by... the Ming Hsieh Foundation; Samsung Healthcare; ... Wright Foundation; and the NIH [grant numbers NIH 5R01CA257610–04, NIH 1R21CA267849–01, and NIH F30CA257401–04].

Contributions (English)

Author Contributions

P.J., J.M.R., and S.M. led the development of segmentation models. P.J., J.M.R., and D.P. led the development of detection models. P.J., J.M.R., and M.M. conducted radiomic analysis. D.H., P.M., and V.D. acquired the radiologic images and performed segmentation and data quality control. J.M.R., X.L., and S.C. performed statistical analysis on clinical applications of the pipeline. P.J. and J.M.R. wrote the first draft of the manuscript. T.T., A.G., A.A., and V.D. oversaw the study. All authors read and approved the final manuscript.

CRediT authorship contribution statement

Duddalwar Vinay: Writing – review & editing, Supervision, Software, Resources, Project administration, Methodology, Investigation, Funding acquisition, Formal analysis, Data curation, Conceptualization. Malewar Shreyas: Writing – review & editing, Visualization, Validation, Methodology, Investigation. Patel Daksh: Writing – review & editing, Methodology. Muellner Matt: Writing – review & editing, Validation. Hwang Darryl: Writing – review & editing, Data curation. Lei Xiaomeng: Visualization, Formal analysis. Cen Steven: Writing – review & editing, Visualization, Formal analysis. Triche Timothy: Writing – review & editing, Supervision, Data curation. Goldkorn Amir: Writing – review & editing, Funding acquisition, Data curation, Conceptualization. Mohamed Passant: Writing – review & editing, Data curation. Oberai Assad: Writing – review & editing, Visualization, Supervision, Resources, Project administration, Methodology, Investigation, Formal analysis, Conceptualization. Pawan S J: Writing – review & editing, Writing – original draft, Visualization, Validation, Methodology, Investigation, Formal analysis, Data curation, Conceptualization. Rich Joseph Matthew: Writing – review & editing, Writing – original draft, Visualization, Validation, Methodology, Investigation, Funding acquisition, Formal analysis, Data curation, Conceptualization.

Data Availability (English)

The data that has been used is confidential.

Conflict of Interest (English)

The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: Joseph Rich reports financial support was provided by Radiological Society of North America. Vinay Duddalwar reports financial support was provided by Ming Hsieh Foundation. Vinay Duddalwar reports financial support was provided by Samsung Healthcare. Vinay Duddalwar reports financial support was provided by Mann Foundation. Amir Goldkorn reports financial support was provided by National Institutes of Health. Amir Goldkorn reports financial support was provided by Wright Foundation. Vinay Duddalwar reports a relationship with Radmetrix that includes: consulting or advisory. Vinay Duddalwar reports a relationship with Westat Inc that includes: consulting or advisory. Vinay Duddalwar reports a relationship with Roche that includes: consulting or advisory. Vinay Duddalwar reports a relationship with DeepTek that includes: consulting or advisory. Amir Goldkorn reports a relationship with OncoSet conference (Northwestern University) that includes: non-financial support. Amir Goldkorn reports a relationship with Roswell Park grant rounds (Roswell Park) that includes: non-financial support. Amir Goldkorn reports a relationship with Norris Comprehensive Cancer Center Executive Committee that includes: board membership. Amir Goldkorn reports a relationship with SWOG GU Executive Committee that includes: board membership. Amir Goldkorn reports a relationship with NIH CONC Study Section that includes: board membership. If there are other authors, they declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Files

1-s2.0-S3050577125000039-main.pdf

Files (3.7 MB)

Name Size Download all
md5:9e2b08be94d0b51b039b8cbaa1e9406b
3.7 MB Preview Download

Additional details

Funding

Radiological Society of North America
Alfred Mann Foundation
National Institutes of Health
5R01CA257610–04
National Institutes of Health
1R21CA267849–01
National Institutes of Health
F30CA257401–04

Dates

Available
2025-02-07
Version of record

Caltech Custom Metadata

Caltech groups
Division of Biology and Biological Engineering (BBE)
Publication Status
Published