Automated analysis of PSMA-PET/CT studies using convolutional neural networks

Abstract

Purpose: Prostate-specific membrane antigen (PSMA) PET/CT has shown to be more sensitive and accurate than conventional imaging. Visual interpretation of the images causes both intra- and inter-reader disagreement and there is therefore a need for objective methods to analyze the images. The aim of this study was to develop an artificial intelligence (AI) tool for PSMA PET/CT and to evaluate the influence of the tool on inter-reader variability. Approach: We have recently trained AI tools to automatically segment organs, detect tumors, and quantify volume and tracer uptake of tumors in PET/CT. The primary prostate gland tumor, bone metastases, and lymph nodes were analyzed in patients with prostate cancer. These studies were based on non-PSMA targeting PET tracers. In this study an AI tool for PSMA PET/CT was developed based on our previous AI tools. Letting three physicians analyze ten PSMA PET/CT studies first without support from the AI tool and at a second occasion with the support of the AI tool assessed the influence of the tool. A two-sided sign test was used to analyze the number of cases with increased and decreased variability with support of the AI tool. Results: The range between the physicians in prostate tumor total lesion uptake (TLU) decreased for all ten patients with AI support (p=0.002) and decreased in bone metastases TLU for nine patients and increased in one patient (p=0.01). Regarding the number of detected lymph nodes the physicians agreed in on average 72% of the lesions without AI support and this number decreased to 65% with AI support. Conclusions: Physicians supported by an AI tool for automated analysis of PSMA-PET/CT studies showed significantly less inter-reader variability in the quantification of primary prostate tumors and bone metastases than when performing a completely manual analysis. A similar effect was not found for lymph node lesions. The tool may facilitate comparisons of studies from different centers, pooling data within multicenter trials and performing meta-analysis. We invite researchers to apply and evaluate our AI tool for their PSMA PET/CT studies. The AI tool is therefore available upon reasonable request for research purposes at www.recomia.org.