The James Buchanan Brady Urological Institute
 
 
 
  Biomarker Research Laboratory          
Home
Image Analysis Technology
Proteomics Technology
Algorithms
Biorepository & Informatics
Staff
Publications
Contact Us


IMAGE ANALYSIS TECHNOLOGY

In the area of digital quantitative Image analysis, we are focused on the applications of methods applying computer-assisted image analysis system that combines a high resolution 3CCD camera and a microscope with excellent optics to capture tissue image information.  Next, a computer with well-engineered imaging software is used to process the images (nuclear pictures or tissue architecture) and then quantitative information about these pictures can be processed to delineate size, shape, DNA content, and chromatin texture or in the case of tissue architecture spatial topography. The nuclear data can be obtained for the cancer cells as well as the adjacent benign areas next to the cancer areas to arrive at the computational solutions for complexities of nuclear structure in order to predict outcomes such as recurrence, progression to metastasis, and likelihood of cancer-specific death. Studies on tissue architecture require software that can quantify spatial topography using fractal dimensions or other approaches.  Further, this computer-assisted microscope technology also permits quantitative analysis of immunohistochemistry (IHC) of molecular biomarkers found in the nucleus, cytoplasm or the membrane of cancer and benign cells. Currently, we have two types of computer-assisted microscopes for such innovative applications, the AutoCyte Pathology Workstation (TriPath Inc.) and CoolScope (Bacus Labs and Nikon).  Additionally, we have access to the BLISS digital imaging system, which allows us to scan tissue microarrays (TMAs) and providing the data in a unique WebSlide format for subsequent analysis by software of the Nikon CoolScope hardware and software seen below.

CURRENT NUCLEAR GRADING TECHNOLOGY


Nikon CoolScope FOR VIRTUAL MICROSCOPY

 

 

Nikon CoolScope SYSTEM:

    1. Performs DNA Ploidy and quantitative nuclear morphometry as the AutoCyte system above.

    2. Generates a nuclear and histologic grade

    3. Quantitates cancer tissue biomarkers stained with antibodies.

    4. Produce quantitative immunohistochemistry (IHC) from WebSlide TMA informatics.

    5. The quantified p300 expression measures intensity & percentage of tumor area positive for the p300 antigen in each PCa case.

NEW TECHNOLOGY Combines BLISS IMAGING SYSTEM + NIKON CoolScope

 As an example, p300 immunohistochemistry (IHC) stained tissues were counterstained with hematoxylin for 1 min and mounted (supplementary Fig. 1). The stained TMAs were scanned with a BLISS virtual slide scanner [Bacus Laboratories, Lombard, IL] at 40X magnification using the WebSlide1 digital microscope slide format. This creates a database input file that lists information on every TMA core and provides an automatic link to the WebSlide1 Net Viewer ActiveX Control for a visual TMA core database. These BLISS virtual slide images are then processed using a TMA score software program using the Nikon CoolScope shown below (Bacus Labs) that quantifies p300 (Histone Acetyltransferase=HAT) expression by measuring percentage of tumor area positive for the p300 antigen in each PCa case.                                                                                           

IHC of P300 and PCA

Non-recurrence P300 Immunohistochemistry Status

Non-recurrence P300 Immunohistochemistry Status

 

The p300/CBP histone acetyltransferase (HAT) causes acetylation of all four core histone residues of the nucleosome where the DNA binds. Modifications of the nucleosome’s net charge by neutralizing the positive charge of lysine ε-amino group alters DNA-histone interactions (cross-talk), which then modify DNA transcriptional activity of the cell. Other nucleosome assembly proteins functionally interact and augment the activity of p300/CBP, and the presence of core histones appears to regulate the interaction between p300 and key nucleosome assembly proteins that establish various chromatin organizational states, impacting nuclear shape and structure and functions (i.e. cell death, cell proliferation, DNA repair etc.).

We continue to utilize the AutoCyteä Pathology Workstation (APW) (See Above) to perform quantitative nuclear morphometry and immunohistochemistry utilizing their QUIC Immuno- and QUIC-DNA software programs.  The QUIC Immuno-software allows the measurement of the amount of a specific tissue biomarker providing a new continuous or categorical input variable to assess multivariately. Also, the AutoCyteä  QUIC- DNA software is also capable of measuring up to 40 cell nuclear shape, size, DNA content, and chromatin organization descriptors of Feulgen-stained prostate epithelia from either cancer or adjacent normal-appearing areas of the prostate gland. We then apply either statistical or Neural Network computer programs to calculate a unique “Morphometric Signature” or Quantitative Nuclear Grade (QNG) solution as another new input biomarker to include in multivariate analysis to predict various PCa outcomes.  Also, we can solve for quantitative nuclear morphometry solutions to assess the role of nuclear structure in cancer therapy or its correlation to cancer biology (i.e. molecular biomarkers). 

As an example, recently the Fisher Biomarker Lab has used quantitative nuclear grade (QNG) to predict biochemical (PSA) recurrence in men with PCa having a 12 year follow-up after their radical prostatectomy. Below we have demonstrated how nuclear morphometry performs compared to pathology to predict PCa biochemical recurrence. The data illustrates a “Morphometric Signature” to predict such recurrence in men with long term follow-up and the results are compared to routine pathology to assess the same outcome. The results in the Table on the left is the “Morphometric Signature” (also referred to as quantitative nuclear grade, QNG) that used 18/40 nuclear descriptors to arrive at a computational solution.  The Receiver Operator Curve (ROC) demonstrates the diagnostic performance of the biomarkers and 1.0 is a perfect ROC score and QNG yielded a 0.87 versus 0.70 for pathology.  The plot on the lower right side of the results is derived from the ROC curve and shows the diagnostic performance for every case studied and notes the sensitivity and specificity of the test.  Below we demonstrate the ability of QNG versus pathology to predict biochemical progression in men following radical prostatectomy. (See Publications.).

MORPHOMETRIC SIGNATURE (QNG) PROGNOSTIC PERFORMANCE

Also, we are studying Gleason grading to determine variations in nuclear structure and eventually trying to quantify changes in glandular architecture of PCa. A graphic illustration of alterations in glandular architecture (cartoon) observed in PCa and an actual microscope photograph of Gleason grade 3, 4 and 5 is demonstrated below.         

 

The future will also use computer-assisted image analysis to quantify tissue architecture such as that illustrated in the pathology slides depicting Gleason Grade (GG) patterns 3, 4, and 5 as they are currently assigned by a pathologist. As GG increases from 3à5, prognosis is more severe.

We were able to characterize alterations size, shape, DNA content, and chromatin texture of each nuclear picture of Gleason Grades (GG) 3, 4 and 5 using the 40 nuclear morphometry features. The nuclear information from accurately Gleason pathologist graded biopsies or radical prostatectomy specimens is used to calculate Quantitative Nuclear Morphometry (QNM) using the best of the 40 nuclear features we can map out similarities and differences to assess heterogeneity of cancer nuclei within Gleason graded tissue samples. Computer-assisted image cytometry was be used to characterize specific nuclear alterations in cancer tissue based upon size, shape, DNA content, and chromatin organization. A QNM signature is able to assess specific changes in such PCa tissue by distinguishing between nuclei of Gleason graded 3, 4, and 5 patterns illustrated above. 

First we compared the differences between adjacent benign (normal) epithelial prostate cells and their adjacent cancer cells for GG 3, 4 and 5.  Clearly as the GG increases the nuclear structure differences become more pronounced as the disease progresses in GG when compared to benign areas.  Next, we compare GG 3, 4 and 5 to each other. The results of the comparisons for GG 3, 4, and 5 is shown in the three Figures below; to the right is demonstrated the variations in nuclear structure between GG3 vs. 4 (black and white bars) and they are quite subtle except for those cell nuclei (bars) shifted to the right of the graph.  In comparing GG3 vs. 5 there are notable differences in nuclear structure (black and white bars) with shifts to the right and left of the graph.  Similarly, a comparison of GG4 vs. 5 demonstrates considerable variations in nuclear structure for the two grades.  Clearly, heterogeneity of nuclear structure is a hallmark of PCa and could be of when value when combining tissue architecture and nuclear changes.


QNM OF GLEASON GRADES 3, 4, and 5 COMPARING BENIGN AND CANCER AREAS


ROC-AUC      0.78

Sensitivity         87%

Specificity        59%

Accuracy         73%

ROC-AUC      0.86

Sensitivity         86%

Specificity        70%

Accuracy         78%

ROC-AUC      0.88

Sensitivity         875%

Specificity        74%

Accuracy         80%


QNM OF GLEASON GRADES 3, 4, and 5 COMPARED TO EACH OTHER

 


Applying Logistic Regression probability plots to GG 3, 4 and 5 nuclei we were able to assess variations in PCa nuclear structure.  The total numbers of cells (nuclei) are on the Y-axes for the sets of comparisons for  GG3 vs GG4 (A), GG3 vs GG4 (B), and GG4 vs GG5 (C) and the % Predictive Probabilities are represented on the X-axes.  Hence, there is a great deal of nuclear heterogeneity within GGs even when a single QNM signature is evaluated.

APPLICATION OF QUANTITATIVE IMMUNOHISTOCHEMISTRY IN PROSTATE CANCER:


Computer-assisted microscopy applied to diagnostic IHC to quantify molecular biomarkers.  As described previously, using the new Nikon-CoolScope in combination with the BLISS Imaging System we have been able to study a nuclear protein (P300) by IHC and the correlation of P300 to alter nuclear structure in PCa.  We also utilized the Nikon CoolScope to quantify P300 in a Tissue Microarray (TMA) obtained from the National Cancer Institute and the PCa outcome was biochemical recurrence.  Using this NCI-TMA we demonstrated that P300 correlates with various size and shape factors in PCa and predicts PSA recurrence.  Also, P300 correlates with PCa Gleason grade and stage.

See the Cox Regression Analysis Table just below and note that Gleason score and P300 predict PSA recurrence and results can be illustrated using a Graphic (Kaplan-Meier) risk analysis below the table. Additionally, below the risk graphic analysis plot are shown “Correlation Networks” that demonstrate how relationships between the P300 IHC and nuclear structure alterations in PCa recurrence versus non-recurrence. Note, that the clear differences demonstrable between the two PCa outcome groups with this approach.  In the future such tools can be used to devise new algorithms to improve the prediction of PCa outcomes.  Further, QNM features of size and shape features will continue to lead us to the importance of genes that control nuclear and tissue architecture in PCa cancer. 

Cox proportional hazards regression


Kaplan-Meier Demonstrating Risk for Prostate Cancer Recurrence based upon Gleason Score and P300


Correlation Networks for Pathology and Morphometry

Correlation network analysis is a graph in which:

  1. A vertex (node) represents an individual nuclear structure feature or nuclear morphometry descriptor (NMD).
  2. An edge (link) between two vertices represents the strength of correlation between the two corresponding NMD features.
  3. Computational algorithms can be developed to visualize and detect underlying regulatory structures among variables over time.

            In our example above; Nuclear Morphometric Descriptors (NMDs) were derived from the PCa tissue areas at the time of radical prostatectomy.  The objective was to predict PSA recurrence using the National Cancer Institute CPCTR-TMA.
 

    1. For each NCI-CPCTR group a threshold of 0.40, an absolute value of correlations, was used to remove insignificant correlation links.
    2. A second threshold of 0.3 was used to remove links with insignificant differences between sample groups.
    3. Finally, nodes without significant links to others in the two groups were removed.
    4. Notably, the results are differentially analyzed to detect differences in correlations that can be potentially linked to PCa recurrence phenotypes. Note in the above networks; the patterns exhibited by the PCa recurrence and non-recurrence groups are quite different.
    5. Next, in the future, the knowledge derived from such correlation networks can be applied to improving algorithm accuracy.


Development of a new image analysis system that can perform accurate Nuclear Roundness Factor measurements on biopsies and prostatectomy specimens.
               
The importance of nuclear morphometry research started at Johns Hopkins University SOM in the 1970’s and eventually resulted in novel technology which accurately measures nuclear roundness in cancer and benign cells.  The nuclear roundness factor represents a dimensionless, size-invariant shape descriptor based on the calculated radius for the measured perimeter divided by the calculated radius for the measured area of the nucleus.  A value 1.0 is a perfect circle and deviation from circularity is reflected by values > 1.0.  For each nucleus analyzed, the computer records the nuclear area, perimeter and nuclear roundness factor.  Nuclear roundness was defined as the degree to which the nucleus in cross section approximated a perfect circle. Circumference (R) is obtained as follows: the distance around a nuclear perimeter can be measured directly, and the computer then calculates R by assuming that the measured perimeter has been obtained from a perfect circle (P=2πR). The area (r) is obtained as follows: the computer, by integration, calculates the actual area of the nucleus and then calculates ‘‘r’’ for an equivalent circle by assuming that the nuclear area measured was contained within a perfect circle (A = πr2).

The nuclear roundness variance (NRV) biomarker though one of the best morphometric predictors of progression-free survival (PFS) (See Figure A below) is currently not available commercially and the image analysis-based assay would have to be re-engineered to more rapidly and in a semi-automated fashion measure nuclear roundness.

Currently, we are discussing with imaging companies the possibilities to develop a product based upon this biomarker parameter.  Also, Zeiss instruments have an image analysis system that accurately measures nuclear roundness as one of several features in its software programs.  Hence, with appropriate funding resources and a corporate partner we hope to continue to pursue this objective in the future. 





© The Johns Hopkins University, The Johns Hopkins Hospital, and Johns Hopkins Health System. All rights reserved.| Disclaimer
Email: webmaster@urology.jhu.edu | 600 North Wolfe Street, Baltimore, Maryland 21287