Search

Reconstructing 3D objects from 2D images is both challenging for our brains and machine learning algorithms. To support this spatial reasoning task, contextual information about the overall shape of an object is critical. However, such information is not captured by established loss terms (e.g. Dice loss). We propose to complement geometrical shape information by including multi-scale topological features, such as connected components, cycles, and voids, in the reconstruction loss. Our method uses cubical complexes to calculate topological features of 3D volume data and employs an optimal transport distance to guide the reconstruction process. This topology-aware loss is fully differentiable, computationally efficient, and can be added to any neural network. We demonstrate the utility of our loss by incorporating it into SHAPR, a model for predicting the 3D cell shape of individual cells based on 2D microscopy images. Using a hybrid loss that leverages both geometrical and topological information of single objects to assess their shape, we find that topological information substantially improves the quality of reconstructions, thus highlighting its ability to extract more relevant features from image datasets., Comment: Accepted at the 25th International Conference on Medical Image Computing and Computer Assisted Intervention (MICCAI)

Nanoparticles delivering drugs, disseminating cancer cells, and red blood cells (RBCs) during splenic filtration must deform and pass through the sub-micrometer and high aspect ratio interstices between the endothelial cells lining blood vessels. The dynamics of passage of particles/cells through these slit-like interstices remain poorly understood because the in vitro reproduction of slits with physiological dimensions in devices compatible with optical microscopy observations requires expensive technologies. Here, novel microfluidic PDMS devices containing high aspect ratio slits with sub-micrometer width are molded on silicon masters using a simple, inexpensive, and highly flexible method combining standard UV lithography and anisotropic wet etching. These devices enabled revealing novel modes of deformations of healthy and diseased RBCs squeezing through splenic-like slits (0.6--2 um 5--10 um 1.6--11 $\mu$m3) under physiological interstitial pressures. At the slit exit, the cytoskeleton of spherocytic RBCs seemed to be detached from the lipid membrane whereas RBCs from healthy donors and patients with sickle cell disease exhibited peculiar tips at their front. These tips disappeared much slower in patients' cells, allowing estimating a threefold increase in RBC cytoplasmic viscosity in sickle cell disease. Measurements of time and rate of RBC sequestration in the slits allowed quantifying the massive trapping of spherocytic RBCs.

How a nano-searcher finds its nano-target is a general problem in non-equilibrium statistical physics. It becomes vital when the searcher is a damaged DNA fragment trying to find its counterpart on the intact homologous chromosome. If the two copies are paired, that intact homologous sequence serves as a template to reconstitute the damaged DNA sequence, enabling the cell to survive without genetic mutations. To succeed, the search must stop only when the perfect homology is found. The biological process that ensures such a genomic integrity is called Homologous Recombination and is promoted by the Recombinase proteins. In this article, we use torque-sensitive magnetic tweezers to measure the free-energy landscape of the human Recombinase hRad51 protein assembled a DNA fragment. Based on our measurements we model the hRad51/DNA complex as an out-of-equilibrium two-state system and provide a thermodynamical description of Homologous Recombination. With this dynamical two-state model, we suggest a mechanism by which the recombinase proteins discriminate between homologous and a non-homologous sequences.