Application value of prostate-specific antigen density combined with multiparametric MRI in early diagnosis of prostate cancer

Abstract

Background

Diagnosis of prostate cancer in the PSA gray zone (4–10 ng/mL) and PI-RADS 3 cases remains challenging. Although multiparametric MRI (mpMRI) is widely used, its diagnostic accuracy is limited by inter-reader variability and the lack of integration with clinical indicators. Prostate-specific antigen density (PSAD) is a valuable risk stratifier, but its optimal combination with mpMRI remains unclear.

Methods

We developed a deep-learning model that integrates PSAD with mpMRI—including T2-weighted imaging (T2WI), diffusion-weighted imaging (DWI), and apparent diffusion coefficient (ADC) maps derived from DWI. A cross-modal attention-guided (CM-AG) fusion module weights the PSAD and mpMRI feature branches. Performance was assessed in the PSA gray zone and the PI-RADS 3 subgroup. Ablation experiments quantified the incremental contributions of PSAD and CM-AG.

Results

The model achieved AUC = 0.89 in the PSA gray-zone cohort and AUC = 0.83 in PI-RADS 3, outperforming single-modality MRI baselines and PI-RADS–based assessment alone (DeLong p < 0.01). In patients with larger prostate volumes, specificity increased by 10.2 %. Ablation results confirmed that both PSAD and CM-AG contributed materially to performance gains.

Conclusion

Fusing PSAD with mpMRI via cross-modal attention improves diagnostic performance, particularly in challenging subgroups (PSA gray zone, PI-RADS 3). This approach may support more consistent risk stratification and earlier detection.