Related Experiment Video
Updated: Nov 7, 2025
A Swin Transformer-Based Model for Thyroid Nodule Detection in Ultrasound Images
Published on: April 21, 2023
Convolutional Neural Networks for Automated PET/CT Detection of Diseased Lymph Node Burden in Patients with Lymphoma
Amy J Weisman1, Minnie W Kieler1, Scott B Perlman1
1Department of Medical Physics, University of Wisconsin-Madison, Madison Wis (A.J.W., R.J.); Department of Radiology, Wisconsin Institutes for Medical Research, University of Wisconsin-Madison, 1111 Highland Ave, Room 1005, Madison, WI 53705 (M.W.K., S.B.P., T.J.B.); Department of Hematology, Rigshospitalet, Copenhagen, Denmark (M.H.); Faculty of Mathematics and Physics, University of Ljubljana, Ljubljana, Slovenia (R.J.); and Department of Radiology and Medical Imaging, University of Virginia, Charlottesville, Va (L.K.).
Insights
Convolutional neural networks (CNNs) show promise in automatically detecting lymphoma-involved lymph nodes on 18F-FDG PET/CT scans, achieving performance comparable to human experts.
Area of Science:
- Medical Imaging
- Artificial Intelligence
- Oncology
Background:
- Lymphoma staging and treatment response assessment rely on accurate detection of involved lymph nodes.
- Manual analysis of 18F-FDG PET/CT scans for lymph node involvement can be time-consuming and subject to inter-observer variability.
Purpose of the Study:
- To develop and evaluate a deep learning model using convolutional neural networks (CNNs) for automated detection of lymph nodes in patients with lymphoma on 18F-FDG PET/CT images.
- To assess the performance of the CNN model in terms of true-positive rate (TPR) and false-positive (FP) findings.
Main Methods:
- A retrospective study involving 90 lymphoma patients with baseline 18F-FDG PET/CT scans.
- Segmentation of involved lymph nodes by a nuclear medicine physician.
- Training an ensemble of 3D patch-based, multiresolution pathway CNNs using fivefold cross-validation.
- Performance evaluation using TPR and FP counts, with comparison to inter-physician agreement in a subset of 20 patients.
Main Results:
- The CNN model achieved an 85% TPR with an average of 4 FP findings per patient across all 90 patients.
- Performance varied across patients, with TPR ranging from 33% to 100%.
- In a comparison with a second physician, the CNN achieved a 90% TPR at 3.7 FP findings per patient, comparable to the second reader's 96% TPR at 3.7 FP findings.
Conclusions:
- An ensemble of 3D CNNs demonstrated a performance in detecting lymph nodes that is nearly comparable to the variability between two expert physicians.
- This study represents a significant first step towards automated PET/CT assessment for lymphoma.
- The findings suggest the potential for AI-driven tools to aid in the interpretation of oncologic imaging.
Purpose:
To automatically detect lymph nodes involved in lymphoma on fluorine 18 (18F) fluorodeoxyglucose (FDG) PET/CT images using convolutional neural networks (CNNs).
Materials And Methods:
In this retrospective study, baseline disease of 90 patients with lymphoma was segmented on 18F-FDG PET/CT images (acquired between 2005 and 2011) by a nuclear medicine physician. An ensemble of three-dimensional patch-based, multiresolution pathway CNNs was trained using fivefold cross-validation. Performance was assessed using the true-positive rate (TPR) and number of false-positive (FP) findings. CNN performance was compared with agreement between physicians by comparing the annotations of a second nuclear medicine physician to the first reader in 20 of the patients. Patient TPR was compared using Wilcoxon signed rank tests.
Results:
Across all 90 patients, a range of 0-61 nodes per patient was detected. At an average of four FP findings per patient, the method achieved a TPR of 85% (923 of 1087 nodes). Performance varied widely across patients (TPR range, 33%-100%; FP range, 0-21 findings). In the 20 patients labeled by both physicians, a range of 1-49 nodes per patient was detected and labeled. The second reader identified 96% (210 of 219) of nodes with an additional 3.7 per patient compared with the first reader. In the same 20 patients, the CNN achieved a 90% (197 of 219) TPR at 3.7 FP findings per patient.
Conclusion:
An ensemble of three-dimensional CNNs detected lymph nodes at a performance nearly comparable to differences between two physicians' annotations. This preliminary study is a first step toward automated PET/CT assessment for lymphoma.© RSNA, 2020.
