Overview of Image Analysis for Prostate MRI
Kirsten Selnæs1

1Department of Radiology and Nuclear Medicine, St. Olavs hospital, Trondheim University Hospital, Trondheim, Norway

Synopsis

Prostate cancer is one of the most frequently diagnosed cancers worldwide and multiparametric MRI is an important part of the diagnostic work-up of prostate cancer patients. Suggestions for standardization of acquisition and interpretation of the images are provided in the PI-RADS v. 2 system, which is increasingly used in clinical practice. New approaches for image analysis, where quantitative features are extracted from the images, allow for the quantitative and multidimensional nature of image data to be exploited. Various machine learning approaches has been used in a wide range of applications within evaluation of prostate mpMRI.

Background

Over the past years, magnetic resonance imaging (MRI) has become an important step in prostate cancer evaluation. A cost effectiveness analysis based on the Prostate MR Imaging Study (PROMIS) indicate that a strategy where MRI is performed first, followed by up to two rounds of MRI-targeted biopsies is both effective and cost effective for the diagnosis of prostate cancer, compared to a strategy with biopsy first [1]. With prostate cancer being one of the most frequently diagnosed cancers worldwide [2], pre-biopsy MRI of all patients with suspected prostate cancer will substantially increase the workload in the radiology department. It is therefore important not only to have effective acquisition protocols, but also to have reliable and time-effective image analysis to secure proper diagnosis for this large patient group.

Current clinical standard

A great work towards standardization of acquisition and interpretation of multi-parametric MRI (mpMRI) of the prostate has resulted in the Prostate Imaging-Reporting and Data System (PI-RADS) v2 [3]. With this framework it has been shown that mpMRI and MRI-directed biopsies provides higher accuracy than systematic TRUS biopsies for diagnosis of prostate cancer [4]. However, image interpretation is still based on manual evaluation by radiologists, which is time consuming, and the radiologist’s training will affect the quality of the results. Further, the PI-RADS v2 system does not include spectroscopic imaging or more advanced image analysis for DCE-MRI, T2WI or DWI, although ongoing research show the potential additional value of such analyses [5-8].

New opportunities and future perspectives

In the era of big data and artificial intelligence, the opportunities to exploit the MR images beyond the individual radiologist’s qualitative evaluation seems endless. With radiomics approaches, where numerous quantitative features are extracted from medical images, the quantitative and multidimensional nature of image data can be exploited [9]. To apply machine learning/radiomics approaches for prostate cancer diagnosis, pre-processing of images, including segmentation of the prostate[10] and co-registration[11] of the various images from mpMRI are required. Quantitative image features typically related to volume, shape, intensity and texture are then extracted from the selected regions in the images. These features are then combined with various clinical data and used to develop diagnostic, predictive or prognostic models for different outcomes [9, 12].

Machine learning has already been used in a wide range of applications within evaluation of prostate mpMRI. It has been shown that machine learning could be used to provide objective risk stratification of prostate cancers based on mpMRI [13]. Bonekamp et al also showed improved performance for differentiation of benign versus malignant prostate lesions for radiomic machine learning and quantitative measurement of mean ADC compared to radiologist’s assessment [14]. For local staging of prostate cancer, it has been shown that extra-prostatic extension can be predicted by using texture analysis and machine learning [15]. In follow-up of patients, it has been shown that features from pretreatment images can predict biochemical recurrence [16], and that radiomics features can be used to categorize risk in patients under active surveillance [17] . Machine learning can also be used to generate tumor probability maps based on multi-parametric and multi-modal prostate imaging to help guide biopsies towards the most aggressive part of the tumor [18].

Another area where automated analyses of prostate mpMRI would be useful is in detection of patients with elevated PSA who do not need biopsies. In a cohort of biopsy-naïve men one can expect that up to 50% have nonsuspicious MRI. A large prospective multicenter study showed that not performing biopsy in patients with nonsuspicious MRI (based on radiologist’s PI-RADS evaluation) would lead to missing clinical significant prostate cancer in only 4% [19]. With the increasing volume of patients referred to pre-biopsy prostate MRI, a reliable and effective tool for automatic detection of cancer-free patients would be time-saving support for the radiologists.

Acknowledgements

No acknowledgement found.

References

1. Faria, R., et al., Optimising the Diagnosis of Prostate Cancer in the Era of Multiparametric Magnetic Resonance Imaging: A Cost-effectiveness Analysis Based on the Prostate MR Imaging Study (PROMIS). Eur Urol, 2018. 73(1): p. 23-30.

2. Bray, F., et al., Global cancer statistics 2018: GLOBOCAN estimates of incidence and mortality worldwide for 36 cancers in 185 countries. CA Cancer J Clin, 2018. 68(6): p. 394-424.

3. Weinreb, J.C., et al., PI-RADS Prostate Imaging - Reporting and Data System: 2015, Version 2. Eur Urol, 2016. 69(1): p. 16-40.

4. Padhani, A.R., et al., Prostate Imaging-Reporting and Data System Steering Committee: PI-RADS v2 Status Update and Future Directions. Eur Urol, 2019. 75(3): p. 385-396.

5. Parra, N.A., et al., Habitats in DCE-MRI to Predict Clinically Significant Prostate Cancers. Tomography, 2019. 5(1): p. 68-76.

6. Wibmer, A., et al., Haralick texture analysis of prostate MRI: utility for differentiating non-cancerous prostate from prostate cancer and differentiating prostate cancers with different Gleason scores. Eur Radiol, 2015. 25(10): p. 2840-50.

7. Brunsing, R.L., et al., Restriction spectrum imaging: An evolving imaging biomarker in prostate MRI. J Magn Reson Imaging, 2017. 45(2): p. 323-336.

8. Mazaheri, Y., et al., Characterization of prostate cancer with MR spectroscopic imaging and diffusion-weighted imaging at 3Tesla. Magn Reson Imaging, 2019. 55: p. 93-102.

9. Gillies, R.J., P.E. Kinahan, and H. Hricak, Radiomics: Images Are More than Pictures, They Are Data. Radiology, 2016. 278(2): p. 563-77.

10. Tian, Z., et al., PSNet: prostate segmentation on MRI based on a convolutional neural network. J Med Imaging (Bellingham), 2018. 5(2): p. 021208.

11. Klein, S., et al., elastix: a toolbox for intensity-based medical image registration. IEEE Trans Med Imaging, 2010. 29(1): p. 196-205.

12. Stoyanova, R., et al., Prostate cancer radiomics and the promise of radiogenomics. Transl Cancer Res, 2016. 5(4): p. 432-447.

13. Varghese, B., et al., Objective risk stratification of prostate cancer using machine learning and radiomics applied to multiparametric magnetic resonance images. Sci Rep, 2019. 9(1): p. 1570.

14. Bonekamp, D., et al., Radiomic Machine Learning for Characterization of Prostate Lesions with MRI: Comparison to ADC Values. Radiology, 2018. 289(1): p. 128-137.

15. Stanzione, A., et al., Detection of Extraprostatic Extension of Cancer on Biparametric MRI Combining Texture Analysis and Machine Learning: Preliminary Results. Acad Radiol, 2019.

16. Shiradkar, R., et al., Radiomic features from pretreatment biparametric MRI predict prostate cancer biochemical recurrence: Preliminary findings. J Magn Reson Imaging, 2018. 48(6): p. 1626-1636.

17. Algohary, A., et al., Radiomic features on MRI enable risk categorization of prostate cancer patients on active surveillance: Preliminary findings. J Magn Reson Imaging, 2018.

18. Elschot, M., et al., Combined (18)F-Fluciclovine PET/MRI Shows Potential for Detection and Characterization of High-Risk Prostate Cancer. J Nucl Med, 2018. 59(5): p. 762-768.

19. van der Leest, M., et al., Head-to-head Comparison of Transrectal Ultrasound-guided Prostate Biopsy Versus Multiparametric Prostate Resonance Imaging with Subsequent Magnetic Resonance-guided Biopsy in Biopsy-naive Men with Elevated Prostate-specific Antigen: A Large Prospective Multicenter Clinical Study. Eur Urol, 2018.

Proc. Intl. Soc. Mag. Reson. Med. 27 (2019)