Interdisciplinary Centre
for Bioinformatics
Home > Projects > Computational Microscopy > 3D Reconstruction of Cervical Carcinoma Invasion Fronts Search  |  Sitemap  |  Imprint  |

LIFE
Bioinformatik Uni Leipzig

3D Reconstruction of Cervical Carcinoma Invasion Fronts


L.-C. Horn
Universitätsklinikum Leipzig
Institut for Pathologie

M. Höckel , R. Scherling , J. Einenkel
Universitätsfrauenklinik

U.-D. Braumann, J.-P. Kuska, D. Drasdo
IZBI

M. Löffler
IMISE


Background and Material


Tumour invasion of squamous epithelial carcinoma (pT1b1) of the uterine cervix varies in different patterns with major implications for prognosis and therapy.

Research Objective


Analysis of tumour morphology by reconstructing 3d tissue data using image processing techniques (registration, segmentation, visualisation, quantification).

3D Reconstructed Tumour Invasion Fronts: Surface Renderings & MPRs


Specimen 13 (500 slices @ 5µm, 89.3mm³) Specimen 6 (300 slices @ 10µm, 104.7mm³) Specimen 9 (150 slices @ 10µm, 100.5mm³)
Specimen 14 Specimen 6 Specimen 9

Main Results


  • very detailed insight into tumour invasion front within the specimen
  • invasion 'per continuitatem', no separated tumour 'islets'
  • invasion patterns form a 'continuum' of compactnesses (see figure 1)
  • compactness basically corresponds to pathologist's assessments
  • achieved the most detailed 3d reconstruction of a solid tumour's invasion front so far

Figure 1: Tumour invasion quantification using discrete compactness


Previous State of the Art: Verbal Tumour Invasion Assessment using 2D Histological Sections


closed moderately dissociated diffuse

In Detail: Our Dedicated Processing Chain


  1. Tissue Specimen (embedded in Paraffin Wax)
  2. Stained Serial Section (H&E)
  3. Digitisation (Transmitted Light Microscopy)
  4. Rigid Registration (Fourier-Mellin Invariant, Phase-Only Matched Filtering)
  5. Colour Adaptation (Multivariate Reference Distribution in RGB)
  6. Polynomial Non-Linear Registration (Least-Squares Estimation, Order: 5)
  7. Staining-Based Tumour Probability (Estimated Multivariate Densities of Normalised Colour Valences)
  8. Curvature-Based Non-Linear Registration (Minimisation of Curvature for the Displacement Vector Field)
  9. Total Variation Filtering (Edge Preserving Smoothing)
  10. Tumour Segmentation (Thresholding)
  11. 3D Tumour Visualisation (Surface Rendering)
  12. Tumour Invasion Quantification (Discrete Compactness)

top