Deep Learning in Prostate Cancer Diagnosis Using Multiparametric Magnetic Resonance Imaging With Whole-Mount Histopathology Referenced Delineations

Abstract

Background: Multiparametric magnetic resonance imaging (mpMRI) plays an important role in the diagnosis of prostate cancer (PCa) in the current clinical setting. However, the performance of mpMRI usually varies based on the experience of the radiologists at different levels; thus, the demand for MRI interpretation warrants further analysis. In this study, we developed a deep learning (DL) model to improve PCa diagnostic ability using mpMRI and whole-mount histopathology data. Methods: A total of 739 patients, including 466 with PCa and 273 without PCa, were enrolled from January 2017 to December 2019. The mpMRI (T2 weighted imaging, diffusion weighted imaging, and apparent diffusion coefficient sequences) data were randomly divided into training (n = 659) and validation datasets (n = 80). According to the whole-mount histopathology, a DL model, including independent segmentation and classification networks, was developed to extract the gland and PCa area for PCa diagnosis. The area under the curve (AUC) were used to evaluate the performance of the prostate classification networks. The proposed DL model was subsequently used in clinical practice (independent test dataset; n = 200), and the PCa detective/diagnostic performance between the DL model and different level radiologists was evaluated based on the sensitivity, specificity, precision, and accuracy. Results: The AUC of the prostate classification network was 0.871 in the validation dataset, and it reached 0.797 using the DL model in the test dataset. Furthermore, the sensitivity, specificity, precision, and accuracy of the DL model for diagnosing PCa in the test dataset were 0.710, 0.690, 0.696, and 0.700, respectively. For the junior radiologist without and with DL model assistance, these values were 0.590, 0.700, 0.663, and 0.645 versus 0.790, 0.720, 0.738, and 0.755, respectively. For the senior radiologist, the values were 0.690, 0.770, 0.750, and 0.730 vs. 0.810, 0.840, 0.835, and 0.825, respectively. The diagnosis made with DL model assistance for radiologists were significantly higher than those without assistance (P < 0.05). Conclusion: The diagnostic performance of DL model is higher than that of junior radiologists and can improve PCa diagnostic accuracy in both junior and senior radiologists.

Keywords: deep learning; detection; magnetic resonance imaging; prostate cancer; segmentation.

Figure 1

Study flowchart of patient selection. PSA, prostate-specific antigen; mpMRI, multiparametric MRI; US, ultrasound.

Figure 2

Flowchart of region of interest delineation for prostate cancer lesion. All the prostate cancer lesions were manually labeled on the magnetic resonance images using whole-mount histopathology as a reference. Representative cases of prostate cancer in different zone distributions: (A) the lesion is in the left peripheral zone, (B) in the right peripheral and transition zone, and (C) in the transition zone and anterior fibromuscular stroma.

Figure 3

Flowchart of the study. The blocks highlighted in blue (prostate gland segmentation network, prostate cancer classification network, prostate cancer segmentation/detection network) denote network models used in our study. “Crop” represents a fixed size region of interest (ROI) to crop the prostate gland according the result of the prostate gland segmentation network. The cropped ROI of ADC and DWI would be registered to the cropped ROI of the T2-weighted imaging (T2WI) and then three cropped ROI would be fed into the prostate cancer classification network. “Positive” represents the positive output of the classification network; in that case, the cropped ROI would be fed into the prostate cancer segmentation network to obtain the lesion region. “Negative” represents the negative output of the classification network; in that case, the cropped ROI would not be fed into the prostate cancer segmentation network.

Figure 4

The graph shows the receiver operating characteristic (ROC) curve for prostate classification network performance. The ROC curves for validation set (A) and test set (B) show area under the curve (AUC) of 0.871 and 0.797, respectively. DOC1, senior radiologist; DOC2, junior radiologist.

Figure 5

Demonstrate representative prostate cancer (PCa) example of radiologists negative (A–C) and deep learning (DL) model positive (D,E). Images show a case of DL model segmentation in a 60-year patient in a test set with prostate-specific antigen (PSA) of 5.59 ng/mL. Axial T2-weighted image (A) shows an ill-defined area of little low signal in the right peripheral zone (arrow), with slight restricted diffusion on apparent diffusion coefficient (ADC) maps (B). (C) Diffusion weighted imaging (DWI) (b-value 1,500 sec/mm²) shows slightly increased signal in this region, with an obvious conspicuity over background normal signal; this lesion would be PI-RADS score 3 for magnetic resonance imaging (MRI). (D,E) show overlapping areas between DL focused PCa region and genuine cancer location. The overlapped areas are colored in red. The software ITK-SNAP was used to open the probability map and MR images at the same time. Through the software function, the probability map is displayed as a jet type color map and overlappedon the T2 weighted imaging (T2WI) to obtain (E); The window width and window level of the probability map is adjusted to 0.5 and 0.75 respectively to display the probability map of the detected cancer area and overlapped on the image to obtain (D).