AI-guided prostate biopsies slightly outperform doctor judgment in cancer detection
A new study of nearly 10,000 men found that software-assisted biopsies detected prostate cancer 3% more often than biopsies guided by radiologist expertise alone. The finding could reshape how hospitals choose biopsy methods and may influence purchasing decisions for imaging software platforms competing in the urology market.
Originaltitel: Comparison of software vs. cognitive-based fusion-targeted biopsies for prostate cancer diagnosis.
Prostatacancer diagnostiseras mer tillförlitligt med mjukvarubaserad fusionsguiderad biopsi än kognitiv fusion. Svenska forskare vid Uppsala universitet jämförde båda metoderna på 10 025 män mellan 2020 och 2024. Mjukvarubaserad teknik detekterade prostatacancer 5 procent oftare än kognitiv fusion — motsvarande 3,3 procentenheter absolut ökning. För aggressiv cancer (Gleason ≥4+3) var skillnaden större: 13 procent högre detektionsfrekvens. Skillnaden blev särskilt markant vid svaga radiologiska fynd (PI-RADS 3), där mjukvara detekterade cancer 12 procent oftare än kognitiv bedömning. Resultaten förändrar taktiken för regionala inköpschefer och chefsläkare. Investeringar i automatiserad fusionsteknologi sannolikt minskar missade cancerfall och reducerar upprepade biopsier. Regulatorisk godkännande för ökad kodning till mjukvarubaserade procedurer bör följas.
OBJECTIVES: This study aimed to compare the effectiveness of software-based fusion-targeted biopsies (STBx) versus cognitive fusion-targeted biopsies (CogTBx) in detecting prostate cancer (PCa) and clinically significant PCa. MATERIALS AND METHODS: Men aged 30-85 were included in a target trial emulation if they had undergone a STBx or CogTBx of Prostate Imaging Reporting and Data System (PI-RADS) 3-5 lesions in 2020-2024. Using log-link binary regression models with inverse probability of treatment weighting to adjust for confounding, we estimated the associations between biopsy technique and detection of PCa and clinically significant PCa (Gleason ≥3 + 4 and Gleason ≥4 + 3) by use of adjusted relative risks (aRR) with 95% confidence intervals (CIs) obtained via bootstrapping. RESULTS: A total of 2092 men who underwent STBx and 7933 men who underwent CogTBx in 2020-2024 were identified in PCBase Xtend. The overall proportion diagnosed with PCa was 64%. STBx detected PCa with a slightly higher frequency (aRR, 1.05; 95% CI, 1.02-1.09), corresponding to an absolute increase of 3.3 percentage points (95% CI 1.2-5.3). The relative risk for detection of Gleason ≥4 + 3 was somewhat larger (aRR, 1.13; 95% CI, 1.03-1.23). The largest relative difference for detection of PCa was observed in PI-RADS 3 lesions (aRR, 1.12; 95% CI, 1.01-1.24), and this pattern became even more apparent when the outcome was restricted to clinically significant PCa. CONCLUSIONS: Software-based fusion-targeted biopsies detected slightly more PCa and clinically significant PCa compared to cognitive fusion-targeted biopsies, particularly in men with PI-RADS 3 lesions. Prioritizing PI-RADS 3 lesions for software-based targeted fusion biopsy could optimize diagnostic effectiveness.