Prostate cancer (PCa) is the most prevalent cancer among the males. PCa detection based on multi-parametric magnetic resonance imaging (mpMRI) can provide precise target points for puncture robots to enhance the accuracy of biopsy procedures. Deep learning (DL) methods have been shown to have better performance than traditional methods on mpMRI-based PCa detection. However, most of the existing DL methods rely on the accurate segmentation of prostate regions, and the calibration of true labels requires time-consuming manual segmentation steps. Meanwhile, the interference of redundant information makes the DL model performance improvement limited. For these reasons, a novel positional-aware attention PCa detection network (PAPDN) is proposed. PAPDN can focus on the position features of PCa lesions and the correlation of mpMRI on channels. It can suppress the interference of redundant information generated by similar structures during PCa detection. The performance of PAPDN is evaluated with the prostate mpMRI dataset collected by Radboud University Medical Center (Radboudumc) in the Netherlands. The results show that PAPDN outperforms other similar algorithms on several rating metrics.