Johns Hopkins Medicine Johns Hopkins Logo
make a gift
hopkins urology on you tube  twitter urology department at hopkins   facebook urology department at hopkins  blog urology department at hopkins  
                
   
Fisher Biomarker Research Laboratory
EVOLUTION OF QUANTITATIVE NUCLEAR MORPHOMETRY
Figure 1 provides a summary of the evolution of technology.

FIGURE 1 David Diamond et al. 1982; nuclear Shape using GraphPad

prostate cancer research

Veltri-Partin Lab Imaging technology 1992-2013.  In order to best interpret the Feulgen stained nuclei, we have used the AutoCyte Pathology Workstation (APW, TriPath Inc., Burlington, NC, USA) fitted with a 3CCD color camera is employed to capture the information based on equations that calculate nuclear size, shape, texture and DNA content with DNA ploidy based on a single-step pixel map of each nucleus. Figure 2 illustrates how the APW operates to generate the quantitative nuclear grader (QNG) variable.

FIGURE 2 – Semi-automated AutoCyte image analysis with Feulgen
 

prostate cancer research

Veltri et al. developed quantitative nuclear grade (QNG) solutions to predict CaP stage, grade, biochemical progression, metastasis and survival (brief summary below):

  1. Veltri RW et al. in 1994, predicted biochemical recurrence in a patient cohort included 115 patients with clinically localized CaP, and the mean follow-up period in 70/115 patients without disease progression was 10.4±1.7 years.  Journal of Cellular Biochemistry, Supplement 19: 249-258, 1994.
  2. Khan MA et al. in 2003 assessed aggressive CaP using nuclear morphometry and predicted progression to metastasis and/or CaP mortality in 227 RP surgical specimens by employing the APW imaging system and applying the QNG analysis. The combined pathology-QNG model retained lymph node status, prostatectomy Gleason score, and QNG, yielding a ROC AUC=86% with an accuracy of 76% at 90% sensitivity. Cancer, 2583-2591, 2003.
  3. Veltri RW et al. in 2004 calculated a nuclear motphometry solution using a tissue microarray (TMA) made from 186 patients (cancer and adjacent benign areas). Clearly, this effort demonstrated a “Field Effect” in the benign areas close to CaP zone.  Clinical Cancer Research, 10: 3465-3473, 2004.
  4. Sumit I et al in 2008, used nuclear morphometry to assess a National Cancer Institute (NCI) Cooperative Prostate Cancer Tissue Resource (CPCTR) tissue microarray of 92 cases with long-term follow-up to demonstrate that the histone acetyltransferase p300 protein predicts biochemical recurrence. The Prostate, 68: 1097-1104, 2008.
  5. Veltri RW et al. in 2010 confirms that Nuclear Roundness Variance (NRV) of nuclei captured in 1992–1993, after 17 year follow-up, proved to be the significant parameter for prediction of recurrence (0.71), metastasis (0.73) and PCa-specific death (0.81).  The Prostate, 70: 1333-1339, 2010.

prostate cancer research
Currently, our Brady Urological Research Institute laboratory is working with Dr. Anant Madabhushi at Case Western Reserve University (CWRU) to apply biomedical engineering technology.  His Lab uses a variety of methods to analyze histomorphometry shown in Table 1 below.  The Aperio scanning microscope is the key to providing 20X – 40X tissue images for study.


Feature Class

Derived attributes

Relevance to histology

Voronoi Tessellation

Number of nodes, number of edges, cyclomatic number, number of triangles, number of k‐walks, spectral radius, eigen exponent, Randic index, area, roundness factor, area disorder, roundness factor homogeneity [74]

Tissue architecture and arrangement of nuclei.

Delaunay Triangulation

Number of nodes, edge length, degree, number of edges, cyclomatic number, number of triangles, number of k‐walks, spectral radius, Eigen exponent, Wiener index, eccentricity, Randic index, fractal dimension [74]

Minimum Spanning Tree

Number of nodes, edge length, degree, number of neighbors, Wiener index, eccentricity, Randic index, Balaban index, fractal dimension [68]

Cell-Graph (local)

Giant Connected Component, eccentricity, number of edges, Connected Component C,D, E [75-79]

Nuclear, Glandular Morphology

Margin spicularity, fractal dimension, height to width ratio, roundness factor, area overlap ratio, area disorder, perimeter, diameter, explicit shape descriptors (medial axis based shape modeling of individual glands) [71,81].

Nuclear and glandular size boundary, appearance

Cell Orientation Entropy / Co-occurring Gland Tensors

Contrast energy, Contrast inverse moment, Contrast average, Contrast variance, Contrast entropy, Intensity average, Intensity variance, Intensity entropy, Entropy, Energy, Correlation, 2 measures of information [73,82]

Second-order descriptors of nuclear orientation in local neighborhoods

As an example,  our collaboration with Case Western Reserve University (CWRU) used an Adaptive Active Contour Model (AdACM) segmentation model, which uses a shape prior to aid in the identification of individual nuclei. The model invokes unique energy terms in and allows for resolving (a) overlaps between nuclei and (b) with significant computational savings. (see Figure 3 below).

AdACM for segmentation of nuclear images     FIGURE 3

prostate cancer research

Figure 3: (a) Original Gleason grade 4 CaP TMA, (b) automated nuclear segmentation by the AdACM scheme, (c) Magnification of a ROI on the TMA reveals that the scheme is able to accurately resolve overlaps between nuclei, and (d) Delaunay triangulation graph obtained by connecting nuclear centers (segmented in (b)).

prostate cancer research Conclusions from Figure 4:
(1) We developed a hybrid ACM method (AdACM) that is both accurate and computationally efficient (seconds) to segment nuclei to quantify CaP phenotype. (2) Morphologic features derived from nuclei segmented via AdACM discriminates different critical prognostic Gleason grade patterns. (3) Ability to distinguish between Gleason Score 3+4 and 4+3 using AdACM nuclear morphology methods is a very clinically    significant finding.  Clearly we show very good discrimination between Gleason grades 3, 4 with a classification accuracy > 85% and Gleason scores.





© 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