Truace
Subscribe

Investigation on Automatic Classification of Bone, Lymph Node, and Local Lesions in CT Images of Prostate Cancer Patients

Accurate assessment of different types of prostate cancer (PCa) lesions is a crucial part of treatment planning. Following a computed tomography (CT) or a positron emission tomography (PET)/CT imaging scan, a significant amount of time is required from the clinicians to detect lesions and characterize them by their anatomical locations. To facilitate the development of new image analysis applications, we investigated the potential of automated classification of PCa lesions between bone metastases, lymph node met…

By the Truvace desk2 min read1 sourcePeer-reviewedTRV-2026-1931· cite & verify
Investigation on Automatic Classification of Bone, Lymph Node, and Local Lesions in CT Images of Prostate Cancer Patients
Gallium PSMA PET scan by Rauscher, Isabel; Maurer, Tobias; Fendler, Wolfgang P.; Sommer, Wieland H.; Schwaiger, Markus; Eiber, Matthias. CC BY 4.0 · https://creativecommons.org/licenses/by/4.0

In brief

Researchers annotated 4949 prostate cancer lesions across PSMA PET/CT scans from 532 patients, then trained traditional machine learning models and convolutional neural networks to label each lesion as bone metastasis, lymph node metastasis, local lesion or other. The features came from CT, lesion coordinates and the TotalSegmentator anatomical segmentation tool.

The work matters because clinicians currently spend considerable time locating and characterizing lesions during treatment planning, and a simple model might support new image analysis tools. Performance was weaker for local lesions on the external test set, and lesions outside the three main categories remain unresolved.

Main points

  1. The dataset covered 4949 annotated lesions from 18F-PSMA PET/CT images of 532 patients, plus an external test set of 30 patients.
  2. Fifteen features per lesion were extracted automatically from CT and TotalSegmentator output.
  3. XGBoost reached 96.0% overall accuracy, the best of the models tested.
  4. Sensitivity in cross-validation was 97.9% for bone, 95.2% for lymph node and 93.1% for local lesions.
  5. On the external test set, local lesion sensitivity fell to 80.0%.

The gain

A gradient-boosted ML model using lesion coordinates, CT and automated anatomical segmentation features classified bone, lymph node and local prostate cancer lesions with 96.0% overall accuracy, which could cut the clinician time spent characterizing lesions.

The rundown

The study compared traditional ML models with CNNs of different dimensionalities, using CT and TotalSegmentator inputs. The aim was to support development of image analysis applications that reduce manual lesion characterization after CT or PET/CT scans.

In five-fold cross-validation on the primary data, XGBoost identified bone metastases, lymph node metastases and local lesions with sensitivities of 97.9±1.2%, 95.2±1.3% and 93.1±2.8%. On the external test set the figures were 96.3%, 94.0% and 80.0%.

The CNNs underperformed the best ML models at distinguishing local lesions. The authors conclude that a simple ML model can classify the three main lesion types, but more research is needed for other PCa lesion categories.

Sources

  1. Peer-reviewedJournal of Imaging Informatics in Medicine2026-10-16

The debate