Visual inspection of histopathology images of stained biopsy tissue by expert pathologists is the standard method for grading of prostate cancer (PCa). However, this process is time-consuming and subject to high inter-observer variability. Machine learning-based methods have the potential to improve efficient throughput of large volumes of slides while decreasing variability, but they are not easy to develop because they require substantial amounts of labeled training data. In this paper, we propose a deep learning-based classification technique and data augmentation methods for accurate grading of PCa in histopathology images in the presence of limited data. Our method combines the predictions of three separate convolutional neural networks (CNNs) that work with different patch sizes. This enables our method to take advantage of the greater amount of contextual information in larger patches as well as greater quantity of smaller patches in the labeled training data. The predictions produced by the three CNNs are combined using a logistic regression model, which is trained separately after the CNN training. To effectively train our models, we propose new data augmentation methods and empirically study their effects on the classification accuracy. The proposed method achieves an accuracy of $\text{92}\%$ in classifying cancerous patches versus benign patches and an accuracy of $\text{86}\%$ in classifying low-grade (i.e., Gleason grade 3) from high-grade (i.e., Gleason grades 4 and 5) patches. The agreement level of our automatic grading method with expert pathologists is within the range of agreement between pathologists. Our experiments indicate that data augmentation is necessary for achieving expert-level performance with deep learning-based methods. A combination of image-space augmentation and feature-space augmentation leads to the best results. Our study shows that well-designed and properly trained deep learning models can achieve PCa Gleason grading accuracy that is comparable to an expert pathologist.

Deep Learning-Based Gleason Grading of Prostate Cancer From Histopathology Images—Role of Multiscale Decision Aggregation and Data Augmentation

Magnetic resonance elastography (MRE) is a medical imaging technique that combines magnetic resonance imaging (MRI) with mechanical vibrations to generate maps of viscoelastic properties of biological tissue. It serves as a non-invasive tool to detect and quantify mechanical changes in tissue structure, which can be symptoms or causes of various diseases. Clinical and research applications of MRE include staging of liver fibrosis, assessment of tumor stiffness and investigation of neurodegenerative diseases.

Magnetic Resonance Elastography: Physical Background and Medical Applications

Magnetic resonance elastography (MRE) in cancer: Technique, analysis, and applications

Assessment of tissue stiffness is desirable for clinicians and researchers, as it is well established that pathophysiological mechanisms often alter the structural properties of tissue. Magnetic resonance elastography (MRE) provides an avenue for measuring tissue stiffness and has a long history of clinical application, including staging liver fibrosis and stratifying breast cancer malignancy. A vital component of MRE consists of the reconstruction algorithms used to derive stiffness from wave-motion images by solving inverse problems. A large range of reconstruction methods have been presented in the literature, with differing computational expense, required user input, underlying physical assumptions, and techniques for numerical evaluation. These differences, in turn, have led to varying accuracy, robustness, and ease of use. While most reconstruction techniques have been validated against in silico or in vitro phantoms, performance with real data is often more challenging, stressing the robustness and assumptions of these algorithms. This article reviews many current MRE reconstruction methods and discusses the aforementioned differences. The material assumptions underlying the methods are developed and various approaches for noise reduction, regularization, and numerical discretization are discussed. Reconstruction methods are categorized by inversion type, underlying assumptions, and their use in human and animal studies. Future directions, such as alternative material assumptions, are also discussed.

https://analyticalsciencejournals.onlinelibrary.wiley.com/doi/pdf/10.1002/nbm.3935

Stiffness reconstruction methods for MR elastography.

https://www.zora.uzh.ch/id/eprint/167289/1/1-s2.0-S1361841517301871-main.pdf

Robust MR elastography stiffness quantification using a localized divergence free finite element reconstruction.

The purpose of this work was to assess trans-perineal prostate magnetic resonance elastography (MRE) for (1) repeatability in phantoms/volunteers and (2) diagnostic power as correlated with histopathology in prostate cancer patients.



The three-dimensional (3D) displacement field was obtained using a fractionally encoded gradient echo sequence using a custom-made transducer. The repeatability of the method was assessed based on three repeat studies and by changing the driving frequency by 3% in studies on a phantom and six healthy volunteers. Subsequently, 11 patients were examined with MRE prior to radical prostatectomy. The areas under the receiver operating characteristic curves were calculated using a windowed voxel-to-voxel approach by comparing the 2D registered slides, masked with the Gleason score.



For the repeatability study, the average intraclass correlation coefficient for elasticity images was 99% for repeat phantom studies, 98% for ±6 Hz phantom studies, 95% for volunteer repeat studies with 2 min acquisition time, 82% for ±2 Hz volunteer studies with 2 min acquisition time and 73% for repeat volunteer studies with 8 min acquisition time. For the patient study, the average elasticity was 8.2 ± 1.7 kPa in the prostate capsule, 7.5 ± 1.9 kPa in the peripheral zone (PZ), 9.7 ± 3.0 kPa in the central gland (CG) and 9.0 ± 3.4 kPa in the transition zone. In the patient study, cancerous tissue with Gleason score at least 3 + 3 was significantly (p < 0.05) different from normal tissue in 10 out of 11 cases with tumors in the PZ, and 6 out of 9 cases with tumors in the CG. However, the overall case-averaged area under the curve was 0.72 in the PZ and 0.67 in the CG.



Cancerous tissue was not always stiffer than normal tissue. The inversion algorithm was sensitive to (i) vibration amplitude and displacement nodes and (ii) misalignment of the 3D wave field due to subject movement. Copyright © 2014 John Wiley & Sons, Ltd.

MR elastography of prostate cancer: quantitative comparison with histopathology and repeatability of methods

The purpose of this work was (1) to develop a magnetic resonance elastography (MRE) system for imaging of the ex vivo human prostate and (2) to assess the diagnostic power of mono-frequency and multi-frequency MRE and diffusion weighted imaging (DWI) alone and combined as correlated with histopathology in a patient study.



An electromagnetic driver was designed specifically for MRE studies in small-bore MR scanners. Ex vivo prostate specimens (post-fixation) of 14 patients who underwent radical prostatectomy were imaged with MRE at 7 T (nine cases had DWI). In six patients, the MRE examination was performed at three frequencies (600, 800, 1000 Hz) to extract the power-law exponent Gamma. The images were registered to wholemount pathology slides marked with the Gleason score. The areas under the receiver-operator-characteristic curves (AUC) were calculated.



The methods were validated in a phantom study and it was demonstrated that (i) the driver does not interfere with the acquisition process and (ii) the driver can generate amplitudes greater than 100 µm for frequencies less than 1 kHz. In the quantitative study, cancerous tissue with Gleason score at least 3 + 3 was distinguished from normal tissue in the peripheral zone (PZ) with an average AUC of 0.75 (Gd), 0.75 (Gl), 0.70 (Gamma-Gd), 0.68 (apparent diffusion coefficient, ADC), and 0.82 (Gd + Gl + ADC). The differentiation between PZ and central gland was modest for Gd (p < 0.07), Gl (p < 0.06) but not significant for Gamma (p < 0.2). A correlation of 0.4 kPa/h was found between the fixation time of the prostate specimen and the stiffness of the tissue, which could affect the diagnostic power results.



DWI and MRE may provide complementary information; in fact MRE performed better than ADC in distinguishing normal from cancerous tissue in some cases. Multi-frequency (Gamma) analysis did not appear to improve the results. However, in light of the effect of tissue fixation, the clinical implication of our results may be inconclusive and more experiments are needed. Copyright © 2014 John Wiley & Sons, Ltd.

/pdf/mr-elastography-and-diffusion-weighted-imaging-of-ex-vivo-33mglzs8dl.pdf

Louis O. Gagnon

Papers

MR elastography of prostate cancer: quantitative comparison with histopathology and repeatability of methods

MR elastography and diffusion-weighted imaging of ex vivo prostate cancer: quantitative comparison to histopathology.