Diagnosis and Localization of Prostate Cancer via Automated Multiparametric MRI Equipped with Artificial Intelligence

Round 1

Reviewer 1 Report

Nowadays, multiparametric-magnetic resonance imaging (mpMRI) represents the most reliable instrumental exam for the prostate study and for the recognition of its malignant neoplasms. The exam is defined multiparametric because many multi-sequential images are acquired and evaluated to stratify the risk of prostate cancer (PCa) into a standardized classification (Prostate Imaging reporting and Data System, PI-RADS). This operation is extremely operator-dependent and time-consuming: for these very reasons, starting from the past few years, to reduce such an interpretational burden and improve productivity in clinical practice, literature is getting rich in artificial intelligence (AI)-based computer-aided diagnosis studies.

Comment to Authors

Authors should be congratulated for the great work and the interesting topic discussed.

This paper aims to develop and validate a new artificial intelligence (AI)-based computer-aided method of prostate cancer localization and shape determination using mpMRI.

The study included 15 PCa patients who underwent radical prostatectomy (RP) and mpMRI in the Hokkaido University Hospital in the period between April ’08 and August ’17. Starting from the assumption cited that a certain number of studies have reported that detection of cancer location by mpMRI enhances the accuracy of targeted prostate biopsy, the three-step automated diagnostic procedure analyzed has been able to detect with an acceptable accuracy cancer localization and shape.

The manuscript is very well-written, easily readable, but it is lacking in several points that would add value to the entire manuscript:

 

Introduction

1. Referred to lines 39-42: a recent review supported the fundamental role of targeted biopsy, complementary with systematic biopsy, in highlighting csPCa because mpMRI-TRUS fusion only targeted biopsy may misdiagnose a small fraction of PCa due to intralesional GS heterogenicity or multifocality. This paper (doi: 1007/s00261-020-02798-8) could improve the quality of contents of the article. Authors should consider to extract information from.
2. Although targeted biopsies, with the enhancing of their accuracy, could take advantage from optimized evaluation of localization and shape, according to what we pointed out at the first point, it would not offer any benefit in terms of unnecessary biopsies avoided. That could be possible with screening algorithms followed by mpMRI and/or biopsy (doi: 10.1097/JU.0000000000001361, PMID: 32897802).
3. Referred to lines 41-52: whereas in this paragraph the paper seems to focus its main goal on the possibility to improve productivity in clinical practice for radiologists, in the following ones, rather, authors seem to demonstrate that they have found an alternative to bypass the shape-estimation problems of other AI-base CAD. Authors should elucidate the main objective of the manuscript.

Materials and methods

4. Referred to Figure 1: images are too small. Authors should set them bigger without losing images sharpness.

Results

5. Referred to Figure 3: even in that case, images are too small.
6. Referred to the entire text: authors should pay attention at the correspondence between lines and their numbers. Not every number match with a written line, but with a blank one.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

Overall, this is an interesting and timely study which attempts to provide useful practical information to reduce radiologists’ burden in interpreting MRI images, which requires multi-sequential interpretations of multi-slice images.

Artificial intelligence (AI)-based, computer-aided diagnosis for MRI, enables more accurate cancer localization and shape estimation and clearly delineates the physiological boundary and anatomical continuity, especially exploiting the collaboration between the two machine-learning techniques analyzed in this study. I think that this work could be improved focusing the attention only on the index lesion (Cecchini S, Castellani D, Fabbietti P, Mazzucchelli R, Montironi R, Cecarini M, Carnevali F, Pierangeli T, Dellabella M, Ravasi E. Combination of Multiparametric Magnetic Resonance Imaging With Elastic-fusion Biopsy Has a High Sensitivity in Detecting Clinically Significant Prostate Cancer in Daily Practice. Clin Genitourin Cancer. 2020 Oct;18(5):e501-e509. doi: 10.1016/j.clgc.2020.02.006) and improving the MRI estimation of tumor location using a shrinkage factor to correct the discrepancies between the radiological and histopathological data (Radtke, J.P.; Schwab, C.; Wolf, M.B.; Freitag, M.T.; Alt, C.D.; Kesch, C.; Popeneciu, I.V.; Huettenbrink, C.; Gasch, C.; Klein, T.; et al. Multiparametric Magnetic Resonance Imaging (MRI) and MRI-Transrectal Ultrasound Fusion Biopsy for Index Tumor Detection: Correlation with Radical Prostatectomy Specimen. Eur. Urol. 2016, 70, 846–853). I also think that an integration with clinical data should be done, checking if the most suspicious lesion at imaging is the one with the highest Gleason score at histopathological examination and demonstrating that a correct localization of the tumor would reduce the length of biopsies needed to achieve Gleason score agreement between biopsy and whole-mount specimen (Fiorentino V, Martini M, Dell'Aquila M, Musarra T, Orticelli E, Larocca LM, Rossi E, Totaro A, Pinto F, Lenci N, Di Paola V, Manfredi R, Bassi PF, Pierconti F. Histopathological Ratios to Predict Gleason Score Agreement between Biopsy and Radical Prostatectomy. Diagnostics (Basel). 2020 Dec 23;11(1):10. doi: 10.3390/diagnostics11010010. PMID: 33374618; PMCID: PMC7822416). I also think that the very small population size and single center evaluation should be mentioned as limitations of this study, even if the data sample size is very large.

Nevertheless, this study proposes an interesting approach to the task and the readability of the paper is enough to read. Concluding, I think that the authors should state that their findings could have a potential clinical impact being exploited to reduce the minimum length of biopsy cores needed to establish a correct Gleason score. Gleason score, in fact, is one of the key parameters to inform a decision of active surveillance vs. treatment, and the results of this work could increase the accuracy of Gleason score assessment.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

Authors should be congratulated for the great contribute to the challenging topic. All future prospective should lead to improve prostate cancer detection reducing investigations number and to create new and better algorithms to properly manage early stage PCa patients, avoiding overdiagnosis and overtreatment. The main concern of histopathological evaluation is the gap between inter- and intra- variability of GS among biopsy specimens. New algorithms and new artificial intelligences would be improved to reduce the burden of a PCa misdiagnosis. The manuscript is well written, the population is well-enrolled, tables and figures are clear but there are several points warrant a mention:

1. Authors should enlighten the basaline characteristics of PCa patients (such as PSA density, biopsy naïve or not, Prostate volume, GS of biopsy specimen and of radical prostatectomy, PIRADS)
2. Are data available on additional features reported in biopsy report, such as cribriform variant or lympho-vascular invasion? How (AI)-based CAD could be helpful for the patients with these histopathological features or in N1? These data must be discussed.
3. How this method differs from the PIRADS prognostic score?
4. Are data available on extra prostatic extension diagnosed by (AI)-based CAD? Authors should read this paper:
5. How this technique differs from the radiomics to localize the PCa? Authors should read this novel paper about the role of radiomics in the localization of Clinically significant PCa. (PMID:31153556)

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

It's ok.

Reviewer 3 Report

Authors should be congratulated for their great contribution to the challenging topic. All future perspective should lead to improve prostate cancer detection reducing investigations number and creating new and better algorithms to properly manage early-stage PCa patients, avoiding overdiagnosis and overtreatment. The main concern of histopathological evaluation is the gap between inter- and intra- variability of GS among biopsy specimens. New algorithms and new artificial intelligence would be improved to reduce the burden of a PCa misdiagnosis. The manuscript is well written, the population is well-enrolled, tables and figures are clear. The adjustments provided by the Authors are clear and meticulous. The manuscript is now suitable for publication.