Artificial intelligence model demonstrates potential in prostate cancer detection
The latest development in AI-assisted medical diagnostics comes from the West German Tumor Center (WTZ) in Münster, who have advanced an AI model to near clinically usable performance for detecting prostate cancer on PSMA-PET/CT images. This progress was published in EJNMMI Physics on August 20.
The new AI model, developed as part of a multicenter study since 2024, has shown significant improvements in the detection and quantification of prostate tumors, lymph nodes, and bone metastases. This follows the initial success of an AI model developed in 2022, which had a sensitivity on par with nuclear medicine physicians for detecting prostate cancer tumors and metastases, but had a higher number of false positive lesions.
To overcome the limitation, the researchers doubled the training dataset to 1,064 patient F-18 PSMA-1007 PET/CT scans. This strategic move has resulted in a significant improvement in the model's performance, with the new AI model's sensitivity for detecting tumors and suspected metastases remaining comparable to that of nuclear medicine physicians.
The false positive rate per patient for suspected lymph node metastases has also decreased from 2.85 to 1.08 for the new model, as reported by the researchers. Moreover, the new AI model has a positive predictive value that significantly improved compared to the old model.
The widespread adoption of PSMA-PET/CT has increased the workload for nuclear medicine departments and physicians. The development of this new AI model aims to support the future integration of AI models into clinical workflows, easing the burden on healthcare professionals and potentially improving the speed and accuracy of diagnoses.
Interestingly, a team in Sweden has also developed an AI model for detecting prostate cancer on PSMA-PET/CT images. While both models show promising results, further research is needed to promote transparency and facilitate other studies.
The researchers have made their AI model freely available to the scientific community at recomia.org. They encourage independent validation of the AI model across diverse clinical settings and imaging protocols to ensure its effectiveness in various scenarios.
Nuclear medicine physicians manually annotated suspected lesions on the images as ground truth for comparing the model's performance. The full study is available for access, offering a comprehensive insight into the development and testing of this new AI model.
