Machine Learning Facilitates Hotspot Classification in PSMA-PET/CT with Nuclear Medicine Specialist Accuracy

Area of Science:

• Nuclear Medicine
• Radiology
• Artificial Intelligence in Medicine
• Oncology

Background:

• Gallium-68 prostate-specific membrane antigen positron emission tomography (68Ga-PSMA-PET) is a key imaging modality for detecting prostate cancer (PC) metastases.
• Accurate interpretation of 68Ga-PSMA-PET images relies on visual discrimination by nuclear medicine (NM) specialists to differentiate malignant from physiologic tracer uptake.
• The integration of machine learning (ML) offers potential for automated, objective analysis to support physicians in image interpretation.

Purpose of the Study:

• To develop and evaluate a machine learning (ML)-based tool for the analysis of 68Ga-PSMA-PET images.
• To compare the diagnostic efficacy of the developed ML tool against human readers (NM specialists).
• To investigate the contribution of radiomic features from both PET and low-dose CT (LDCT) data for improved classification accuracy.

Main Methods:

• Five distinct ML algorithms were assessed using multiple positron emission tomography/computed tomography (PET/CT) datasets.
• Radiomic features, including 40 textural features, were extracted from both 68Ga-PSMA-PET and LDCT images.
• A total of 2419 hotspots from 72 patients were analyzed, and ML performance was compared to expert NM readers.

Main Results:

• ML-based analysis achieved high performance metrics, including up to 98% area under the curve (AUC), 94% sensitivity (SE), and 89% specificity (SP), comparable to human readers.
• The incorporation of textural features from native LDCT significantly enhanced classification accuracy.
• Multimodal ML analysis, integrating both PET and LDCT features, demonstrated superior performance, highlighting the importance of morphological features from LDCT.

Conclusions:

• Machine learning models utilizing 68Ga-PSMA-PET/CT radiomics features can accurately classify hotspots with precision comparable to experienced NM physicians.
• The combination of PET and LDCT data in a multimodal ML approach significantly improves diagnostic performance.
• Radiomic and morphological features extracted from LDCT play a crucial role in enhancing the accuracy of ML-based image analysis for prostate cancer detection.