Skip to Main content Skip to Navigation

3D motion estimation and assessment in fluorescence microscopy volume sequences

Abstract : The thesis work deals with the computation and the assessment of 3D motion fields in 3D fluorescence microscopy image sequences. We have investigated 3D matching and variational methods for 3D flow field estimation between two consecutive volumes. For matching, we have developed two original 3D extensions of PatchMatch both involving the discrete Census similarity measure: a super-pixel based method that proceeds slice by slice, and a coarse-to-fine method directly applied to the volumes. We have also designed a protrusion segmentation method on the cell surface along with a matching stage relying on a triangular mesh-based representation. Regarding the dense estimation of 3D flow fields, we have adopted a variational approach, while exploiting the continuous Census signature of voxels in the data term. We have tested three regularization terms: L2, L1, and TV-based regularization. We have also combined the 3D PatchMatch method with the variational method to be able to handle simultaneously large and small motion magnitude. For visual assessment, we have proposed three different color-coded visualization techniques of 3D flow fields. They offer 2D summaries of the 3D flow field, respectively, slice-by-slice, with tri-planar projections, and after maximum intensity point projection. In addition, we have defined a new quantitative error measure for assessing the accuracy of the estimated flow field when no ground truth is available. It involves the angular difference between the local principal orientation of the original source volume and the corresponding one in the volume backward-warped with the 3D computed flow field. We have tested our methods on real microscopy image sequences containing MV3 melanoma cells in collagen environment. When comparing with the state-of-the-art method of Amat et al., and our 3D extension of the classical Horn-and-Schunck method, we found our proposed methods to be the best performing ones.
Complete list of metadatas

Cited literature [241 references]  Display  Hide  Download
Contributor : Abes Star :  Contact
Submitted on : Friday, July 3, 2020 - 11:03:39 AM
Last modification on : Wednesday, September 9, 2020 - 4:15:49 AM
Long-term archiving on: : Thursday, September 24, 2020 - 7:17:00 AM


Version validated by the jury (STAR)


  • HAL Id : tel-02888650, version 1



Sandeep Manandhar. 3D motion estimation and assessment in fluorescence microscopy volume sequences. Medical Imaging. Université Rennes 1, 2019. English. ⟨NNT : 2019REN1S096⟩. ⟨tel-02888650⟩



Record views


Files downloads