PSP net-based automatic segmentation network model for prostate magnetic resonance imaging

https://doi.org/10.1016/j.cmpb.2021.106211Get rights and content

Highlights

  • The CLAHE algorithm is used to enhance the data set.

  • A prostate MRI segmentation model based on PSP Net is proposed.

  • PSP Net model compares segmentation accuracy with FCN and U-Net.

  • PSP Net has the highest segmentation accuracy rate of 0.9865.

  • The AUC of PSP Net is 0.9427 and the ROC curve is closest to the upper left corner.

Abstract

Purpose: Prostate cancer is a common cancer. To improve the accuracy of early diagnosis, we propose a prostate Magnetic Resonance Imaging (MRI) segmentation model based on Pyramid Scene Parsing Network (PSP Net).

Method: A total of 270 prostate MRI images were collected, and the data set was divided. Contrast limited adaptive histogram equalization (CLAHE) was enhanced in this study. We use the prostate MRI segmentation model based on PSP net, and use segmentation accuracy, under segmentation rate, over segmentation rate and receiver operating characteristic (ROC) curve evaluation index to compare the segmentation effect based on FCN and U-Net.

Results: PSP net has the highest segmentation accuracy of 0.9865, over segmentation rate of 0.0023, under segmentation rate of 0.1111, which is less than FCN and U-Net. The ROC curve of PSP net is closest to the upper left corner, AUC is 0.9427, larger than FCN and U-Net.

Conclusion: This paper proves through a large number of experimental results that the prostate MRI automatic segmentation network model based on PSP Net is able to improve the accuracy of segmentation, relieve the workload of doctors, and is worthy of further clinical promotion.

Introduction

The prostate located between the bladder and the urogenital diaphragm and surrounds the root of the urethra. Its shape and size are similar to those of a flat chestnut [1], [2], [3], [4]. With the continuous enhancement of people's living and changes in living habits, the incidence and mortality of prostate cancer have also increased year by year. Statistics from the National Cancer Institute of the United States in 2018 show that 16,490 new prostate cancer patients are expected in 2018, the highest proportion of all male cancers [5]. The data shows that the incidence of prostate cancer [6], [7] is increasing year by year, and it has become the second largest male cancer after lung cancer.

At present, the examination methods for prostate cancer include digital rectal examination [8,9], needle biopsy [10,11], Prostate Specific Antigen (PSA) examination [12,13], Transrectal ultrasonography (TRUS) [14,15], Computed Tomography (CT) [16,17] and Magnetic Resonance Imaging (MRI) [18,19]. MRI is currently the most effective method for diagnosing prostate cancer. Not only can it clearly show that the prostate tumor invades other tissues, but it can also observe the lesion from multiple angles and clearly show the shape of the pelvic bones. Compared with other imaging methods, MRI can effectively help doctors determine the location of prostate cancer, enable doctors to accurately radiotherapy prostate cancer lesions, and effectively avoid damage to other healthy tissues.

In recent years, convolutional neural network (CNN) has made a major breakthrough, constantly improving the accuracy of medical image segmentation. In the field of prostate image segmentation, milletari [20] et al. A 3D version of U-Net v-net is proposed and applied to prostate MRI segmentation for the first time.

Yu [21] et al. proposed three-dimensional neural network architecture with mixed residual connections for prostate MRI segmentation. The architecture effectively utilizes the feature multiplexing characteristics of the residual module and the long-short jump connection, and directly improves the segmentation accuracy of the prostate area at the three-dimensional level. Tian [22] et al. adopted an optimized FCN structure and achieved good prostate segmentation results. Zhu [23] and others proposed a bidirectional recursive deep neural network for MRI prostate image segmentation. In addition to using embedded context features, this method also processes prostate slices as a data sequence and uses gap context to assist segmentation.

Traditional methods have solved the problem of prostate MRI segmentation to a certain extent. However, due to the great variability of prostate tissue in prostate contour, size and shape, and the interference of surrounding tissue and imaging artifacts, automatic segmentation of prostate MRI is still a huge challenge. In our research, we deeply studied CNN algorithm, and proposed a prostate MRI segmentation model based on PSP net.

Section snippets

1.1 Overview of PSP Net model

Deep learning was proposed by Hinton Equals in 2006 [24]. It is a field of machine learning. Its purpose is to learn data features by constructing a network model containing multiple hidden layers and training a large amount of data to improve classification or prediction. The network model used in this study is the PSP Net model, and the discriminant model of prostate cancer is established by fine-tuning the model framework Fig. 1.

Calculation process

Taking the first convolutional layer as an example, the

Enhanced results

The detailed results are shown in Fig. 7. It can be seen that the contrast between the prostate tissue and the surrounding tissues in the original MRI is low, and the dynamic range of the gray scale is small. After the contrast-limited adaptive histogram equalization (CLAHE) is changed, the contrast between the prostate tissue and the surrounding tissues is enhanced, and the gray range of the image is enlarged, the details in the image are clearer, and there is no phenomenon such as blurring

Discussion

Prostate tissue segmentation is a basic task for computer-assisted doctors to diagnose the condition, and it is also one of the tasks with strong challenges and potential research value in the field of medical image processing [29], [30], [31]. Before confirming the patient's condition, the doctor needs to accurately segment the patient's prostate tissue area to determine the degree of disease. Prostate image segmentation has great potential value in the field of medical assisted diagnosis.

Conclusion

Artificial intelligence technology based on deep learning has achieved rapid development in recent years, especially convolutional neural networks for images, which have now played an important role in computer-aided diagnosis systems. Multi-parameter magnetic resonance imaging has the characteristics of multiple dimensions and multiple complexity. It takes a long time for doctors to read the film. At the same time, the accuracy of diagnosis is also very dependent on the experience of

Declaration of Competing Interest

The authors declare no conflicts of interest.

Acknowledgment

Special appreciation is extended to radiologists of the Fifth Affiliated Hospital of Southern Medical University of Guangzhou for their help in imaging the prostate based on the patients they collected.

References (42)

  • F. Shahrokh et al.

    Survivin expression is associated with features of biologically aggressive prostate carcinoma

    Cancer

    (2004)
  • S. Ashida

    Molecular features of the transition from prostatic intraepithelial neoplasia (PIN) to prostate cancer: genome-wide gene-expression profiles of prostate cancers and PINs

    Cancer Res.

    (2004)
  • N.L. Keating et al.

    Variations in hospice use among cancer patients

    J. Natl. Cancer Inst.

    (2021)
  • R. Dreicer et al.

    Sipuleucel-t immunotherapy for castration-resistant prostate cancer

    New Engl. J. Med.

    (2010)
  • M. Chan J

    Plasma insulin-like growth factor-i and Prostate cancer risk: a prospective study

    Science

    (2015)
  • K.E. Richert-Boe et al.

    Screening digital rectal examination and prostate cancer mortality: a case-control study

    J Med Screen

    (1998)
  • FH. Schrioder

    Evaluation of the digital rectal examination as a screening test for prostate cancer

    J. Natl. Cancer Inst.

    (1998)
  • F. Aragona P

    Saturation prostate needle biopsy and prostate cancer detection at initial and repeat evaluation

    Urology

    (2007)
  • S.R. Dubey, S. Chakraborty, Soumendu, Average biased ReLU based cNN descriptor for improved face retrieval. (2018)...
  • S.C. Brö et al.

    Comparative study of two different TRUS-guided sextant biopsy techniques in detecting prostate cancer in one biopsy session

    Eur. Urol.

    (2000)
  • L. Wolfgang et al.

    Computed tomography/magnetic resonance based volume changes of the primary tumour in patients with prostate cancer with or without androgen deprivation

    Radiother Oncol.

    (2000)
  • Cited by (33)

    • Joint task semi-supervised semantic segmentation for TRUS image

      2024, Biomedical Signal Processing and Control
    • Noncontact measurement of tire deformation based on computer vision and Tire-Net semantic segmentation

      2023, Measurement: Journal of the International Measurement Confederation
    View all citing articles on Scopus
    View full text