Semi-Automated Smart Detection of Prostate Cancer using
Machine Learning and a Novel Near-Microscopic Imaging Platform

Daniel Rohrbach- , Jonathan Mamou and Ernest Feleppa
Lizzi Center for Biomedical Engineering, Riverside Research
New York, NY, USA, 10038

Brian Wodlinger and Jerrold Wen
Exact Imaging, Markham
Ontario, Canada, L3R 2N2


Popular version of paper 2pBA3, “Quantitative-ultrasound-based prostate-cancer imaging by means of a novel micro-ultrasound scanner”

Presented Tuesday, December 05, 2017, 1:45-2:00 PM, Balcony M

174th ASA meeting, New Orleans


Prostate cancer is the second-leading cause of male cancer-related death in the U.S. with approximately 1 in 7 men being diagnosed with prostate cancer during their lifetime[i].  Detection and diagnosis of this significant disease presents a major clinical challenge because the current standard-of-care imaging method, conventional transrectal ultrasound, cannot reliably distinguish cancerous from non-cancerous prostate tissue.  Therefore, prostate biopsies for definitively diagnosing cancer are currently delivered in a systematic but “blind” pattern.  Other imaging methods, such as MRI, have been investigated for guiding biopsies, but MRI involves complicated procedures, is costly, is poorly tolerated by most patients, and  demonstrates significant variability among clinical sites and practitioners.  Our study investigated sophisticated tissue-typing algorithms for possible use in a novel, fine-resolution, ultrasound instrument called the ExactVu™ micro-ultrasound instrument by Exact Imaging, Markham, Ontario.  The ExactVu recently has been approved for commercial sale in North America and Europe.  The term micro-ultrasound refers to the near-microscopic resolution of the device.  This new, fine-resolution instrument allows clinicians to visualize previously unseen features of the prostate in real time and enables them to differentiate suspicious regions of the prostate so that they can “target” biopsies to those suspicious regions.  To enable more-objective interpretation of tissue features made visible by the ExactVu, a cancer-risk-identification protocol – called PRI-MUS™ (prostate risk Identification using micro-ultrasound)[ii] – has been developed and validated to distinguish benign prostate tissue from tissue that has a high probability of being cancerous based on its appearance in a micro-ultrasound image.

The paper, “High-frequency quantitative ultrasound for prostate-cancer imaging using a novel micro-ultrasound scanner, which is being presented at the 174th Acoustical Society of America, shows promising results from a collaborative research project undertaken by Riverside Research, a leading biomedical research institution in New York, NY, and Exact Imaging.  The paper describes an approach that successfully applies a combination of (1) sophisticated ultrasound signal processing methods known as quantitative ultrasound and (2) machine-learning and artificial intelligence to analysis of fine-resolution data acquired with the novel micro-ultrasound imaging platform to automate detection of cancerous tissue in the prostate.  Results of the study were very encouraging and showed a promising ability of the methods to distinguish cancerous from non-cancerous prostate tissue.

A database of 12,000 fine-resolution, micro-ultrasound images and correlated biopsy histology has been developed.  The new algorithm for automated detection continues to evolve and is applied to this growing data set.

Future clinical application of the algorithms implemented in the ExactVu would involve scanning a patient with indications of prostate cancer (e.g., as a result of a transrectal palpation or a high level of prostate-specific antigen in the blood) to identify regions of the gland that are sufficiently suspicious for cancer to warrant a biopsy.  As the scan proceeds, the algorithm continuously analyzes the ultrasound signals and automatically indicates to the examining urologist any regions that have a significant risk of being cancerous.  The urologist evaluates the indicated region and makes a clinical judgement on whether the region in fact warrants a biopsy.

The results of this study show an encouraging ability of ultrasound-signal processing and the machine-learning algorithm together with the novel micro-ultrasound instrumentation to depict regions of the prostate that are cancerous with high reliability.  The study demonstrates a promising potential of the algorithms and micro-ultrasound to improve targeting of biopsies, to increase cancer-detection rates, to avoid unnecessary biopsies and associated risks, to support focal therapy more effectively, and consequently to achieve better patient outcomes.


Referenced abstract:
High-frequency quantitative ultrasound for prostate-cancer imaging using a novel micro-ultrasound scanner

[i] American Cancer Association:!/

[ii] Ghai S, et al: Assessing Cancer Risk on Novel 29 MHz Micro-Ultrasound Images of the Prostate: Creation of the Micro-Ultrasound Protocol for Prostate Risk Identification. J. Urol. 2016; 196: 562–569.

Share This