Skip to main page content
U.S. flag

An official website of the United States government

Dot gov

The .gov means it’s official.
Federal government websites often end in .gov or .mil. Before sharing sensitive information, make sure you’re on a federal government site.

Https

The site is secure.
The https:// ensures that you are connecting to the official website and that any information you provide is encrypted and transmitted securely.

Access keys NCBI Homepage MyNCBI Homepage Main Content Main Navigation
. 2021 Dec 6;13(23):6138.
doi: 10.3390/cancers13236138.

AutoProstate: Towards Automated Reporting of Prostate MRI for Prostate Cancer Assessment Using Deep Learning

Pritesh Mehta  1   2 , Michela Antonelli  2 , Saurabh Singh  3 , Natalia Grondecka  4 , Edward W Johnston  5 , Hashim U Ahmed  6 , Mark Emberton  7 , Shonit Punwani  3 , Sébastien Ourselin  2
Affiliations

AutoProstate: Towards Automated Reporting of Prostate MRI for Prostate Cancer Assessment Using Deep Learning

Pritesh Mehta et al. Cancers (Basel). .

Abstract

Multiparametric magnetic resonance imaging (mpMRI) of the prostate is used by radiologists to identify, score, and stage abnormalities that may correspond to clinically significant prostate cancer (CSPCa). Automatic assessment of prostate mpMRI using artificial intelligence algorithms may facilitate a reduction in missed cancers and unnecessary biopsies, an increase in inter-observer agreement between radiologists, and an improvement in reporting quality. In this work, we introduce AutoProstate, a deep learning-powered framework for automatic MRI-based prostate cancer assessment. AutoProstate comprises of three modules: Zone-Segmenter, CSPCa-Segmenter, and Report-Generator. Zone-Segmenter segments the prostatic zones on T2-weighted imaging, CSPCa-Segmenter detects and segments CSPCa lesions using biparametric MRI, and Report-Generator generates an automatic web-based report containing four sections: Patient Details, Prostate Size and PSA Density, Clinically Significant Lesion Candidates, and Findings Summary. In our experiment, AutoProstate was trained using the publicly available PROSTATEx dataset, and externally validated using the PICTURE dataset. Moreover, the performance of AutoProstate was compared to the performance of an experienced radiologist who prospectively read PICTURE dataset cases. In comparison to the radiologist, AutoProstate showed statistically significant improvements in prostate volume and prostate-specific antigen density estimation. Furthermore, AutoProstate matched the CSPCa lesion detection sensitivity of the radiologist, which is paramount, but produced more false positive detections.

Keywords: automatic report; computer-aided diagnosis; convolutional neural network; deep learning; lesion classification; lesion detection; magnetic resonance imaging; prostate cancer; segmentation.

PubMed Disclaimer

Conflict of interest statement

H.U.A. is a paid consultant to Boston Scientific for teaching and training on Rezum for benign prostate hyperplasia treatment and cryotherapy for prostate cancer treatment and is paid for teaching and proctoring HIFU for treating prostate cancer. M.E. receives honoraria from consulting, educational activities, and training from: Sonacare Inc.; NINA Medical; and Angiodynamics Inc. All other authors declare no conflicts of interest.

Figures

Figure 1
Figure 1
AutoProstate framework diagram. AutoProstate consists of three modules: Zone-Segmenter (green), CSPCa-Segmenter (blue), and Report-Generator (purple); solid boxes correspond to module computations, while dashed boxes correspond to module outputs. Yellow boxes indicate AutoProstate inputs from external sources.
Figure 2
Figure 2
AutoProstate Report template, where xx denotes an automatically populated field.
Figure 3
Figure 3
AutoProstate external validation analysis of whole-prostate and zonal segmentations, prostate size measurements, and PSAd, using 80 patients from the PICTURE dataset for which ground-truth segmentations were available: (a) Distribution of Dice coefficients for PZ, CG, and whole-prostate segmentation; (b) Distribution of Abs%Err for transverse, anterior–posterior, and cranio–caudal lengths; (c) Distribution of Abs%Err for PZ and CG volumes; (d) Distribution of Abs%Err for whole-prostate volume estimations by AutoProstate and the experienced radiologist; and (e) Distribution of AbsErr for PSAd calculated by AutoProstate and the experienced radiologist; the ground-truth PSAd value used to compute the AbsErr for AutoProstate and the experienced radiologist was calculated by dividing PSA by the whole-prostate volume derived from the ground-truth whole-prostate segmentation.
Figure 4
Figure 4
PICTURE dataset CSPCa lesion detection (a) ROC curves and (b) PR curves, corresponding to the experienced radiologist (gold) and AutoProstate (blue).
Figure 5
Figure 5
PICTURE dataset axial T2WI, ADC map, Cb2000 DWI, ground-truth lesion contour overlaid on T2WI, probability map overlaid on T2WI, and segmentation overlaid on T2WI: (a) 79-year-old man, PSA 12.57 ng/mL, midgland PZ GS 4 + 3 lesion, Likert 5, AutoProstate probability of CSPCa 100%; (b) 66-year-old man, PSA 7.50 ng/mL, midgland PZ GS 3 + 4 lesion, Likert 3, AutoProstate probability of CSPCa 65%; (c) 64-year-old man, PSA 10.53 ng/mL, apex CG GS 3 + 4 lesion, Likert 5, AutoProstate probability of CSPCa 95%; (d) 56-year-old man, PSA 7.91 ng/mL, base CG GS 3 + 4 lesion, Likert 4, AutoProstate probability of CSPCa 66%; (e) 60-year-old man with stable rectal gas-induced magnetic susceptibility artefact on DWI, PSA 6.15 ng/mL, midgland PZ GS 3 + 4 lesion, Likert 5, AutoProstate probability of CSPCa 88%; and (f) 73-year-old man with bowel peristalsis-induced magnetic susceptibility artifact, PSA 4.09 ng/mL, midgland PZ GS 3 + 4 lesion, Likert 5, AutoProstate probability of CSPCa 49%.
Figure 6
Figure 6
AutoProstate Report for a 64-year-old man with PSA equal to 10.53 ng/mL who participated in the PICTURE study. LESION 1 (probability of CSPCa equal to 95%) corresponds to a biopsy-proven GS 3+4 lesion, while LESION 2 and LESION 3 (probabilities of CSPCa equal to 46% and 7%, respectively) are false positives.
See this image and copyright information in PMC

Similar articles

See all similar articles

Cited by

See all "Cited by" articles

References

    1. Ahmed H.U., El-Shater Bosaily A., Brown L.C., Gabe R., Kaplan R., Parmar M.K., Collaco-Moraes Y., Ward K., Hindley R.G., Freeman A., et al. Diagnostic accuracy of multi-parametric MRI and TRUS biopsy in prostate cancer (PROMIS): A paired validating confirmatory study. Lancet. 2017;389:815–822. doi: 10.1016/S0140-6736(16)32401-1. - DOI - PubMed
    1. Brembilla G., Dell’Oglio P., Stabile A., Damascelli A., Brunetti L., Ravelli S., Cristel G., Schiani E., Venturini E., Grippaldi D., et al. Interreader variability in prostate MRI reporting using Prostate Imaging Reporting and Data System version 2.1. Eur. Radiol. 2020;30:3383–3392. doi: 10.1007/s00330-019-06654-2. - DOI - PubMed
    1. Stanzione A., Ponsiglione A., Di Fiore G.A., Picchi S.G., Di Stasi M., Verde F., Petretta M., Imbriaco M., Cuocolo R. Prostate Volume Estimation on MRI: Accuracy and Effects of Ellipsoid and Bullet-Shaped Measurements on PSA Density. Acad. Radiol. 2021;28:e219–e226. doi: 10.1016/j.acra.2020.05.014. - DOI - PubMed
    1. Distler F.A., Radtke J.P., Bonekamp D., Kesch C., Schlemmer H.-P., Wieczorek K., Kirchner M., Pahernik S., Hohenfellner M., Hadaschik B.A. The Value of PSA Density in Combination with PI-RADSTM for the Accuracy of Prostate Cancer Prediction. J. Urol. 2017;198:575–582. doi: 10.1016/j.juro.2017.03.130. - DOI - PubMed
    1. Yang X., Lei Y., Wang T., Jiang X., Jani A., Mao H., Curran W., Patel P., Liu T., Wang B. 3D prostate segmentation in MR image using 3D deeply supervised convolutional neural networks. Med. Phys. 2018;45:e582–e583.