• Prostate cancer diagnosis with the aid of deep learning in multi-parametric magnetic resonance images
  • Abolfazl Koozari,1,* Mohammadreza Elhaie,2 Iraj Abedi,3
    1. Department of Medical Physics, School of Medicine, Ahvaz Jundishapur University of Medical Sciences
    2. Department of Medical Physics, School of Medicine, Isfahan University of Medical Sciences
    3. Department of Medical Physics, School of Medicine, Isfahan University of Medical Sciences


  • Introduction: Prostate cancer is the second most common malignancy among men globally. Magnetic resonance imaging (MRI) is a useful method in prostate cancer detection. MRI has some benefits such as being a non-ionizing and highly sensitive approach. multiparametric MRI(mpMRI) protocols for prostate cancer detection Aim to increase sensitivity and specificity by compounding anatomical sequences of T1-weighted images (T1WI) and multiplanar T2-weighted images (T2WI) with functional sequences of diffusion-weighted imaging (DWI) and dynamic contrast enhancement (DCE)MRI. mpMRI is used to decrease needless biopsies. mpMRI shows high accuracy and specificity for recognizing clinically significant prostate can r. However, the diagnosis of prostate cancer via mpMRI highly depends on radiologists' expertise. In recent years, Deep learning has been designed for a wide range of applications in medical imaging. Therefore this review aims to investigate the abilities of deep learning methods in diagnosing prostate cancer using mpMRI.
  • Methods: This search was conducted in the Google Scholar database with the following keywords: “prostate cancer” in the title and” diagnosis” and “deep learning” and “magnetic resonance imaging” or “MRI” or “mpMRI” in all fields. We limited the publication time to after 2022 to evaluate the most recent literature. We also used the PubMed database for extra literature searches. In addition, relevant works published on the mentioned scientific websites were investigated. After screening the abstracts, we selected the relevant articles for this study.
  • Results: The total number of papers obtained through the search was 517. We limited our results to 39 papers based on the inclusion values. Among deep learning tactics, convolution neural network (CNN) is the most potent network which is skilled in extracting strong features from the input images that contemplate features from the low to the high level. there are other pre-trained deep learning models such as MobileNetV2, ResNet50V2, Resnet101V2, Resnet152V2, Xception, InceptionResNetV2, and InceptionV3 which could be valuable to detect prostate cancer from given image groups. Many studies showed that mpMRI has a high sensitivity in detecting prostate cancer (more than 85%) also in many studies, it has been said that the diagnostic accuracy of detecting prostate lesions has increased with the help of deep learning.
  • Conclusion: Deep learning has the potential to improve diagnostic accuracy and reduce subjective decision-making in the process of cancer prediction. Deep learning has the “learn” ability, this ability is obtained through the analysis of features and complex image data and structures. The reviewed articles showed that the Deep learning techniques applied to mpMRI, seem to be an effective assistant in predicting and detecting prostate cancer lesions. This paper reports that the ability of deep learning to play a complementary role could aid radiologists in better diagnosis of prostate lesions.
  • Keywords: prostate cancer, magnetic resonance imaging, deep learning, diagnosis