Impact of rescanning and normalization on convolutional neural network performance in multi-center, whole-slide classification of prostate cancer.

Algorithms can improve the objectivity and efficiency of histopathologic slide analysis. In this paper, we investigated the impact of scanning systems (scanners) and cycle-GAN-based normalization on algorithm performance, by comparing different deep learning models to automatically detect prostate cancer in whole-slide images. Specifically, we compare U-Net, DenseNet and EfficientNet. Models were developed on a multi-center cohort with 582 WSIs and subsequently evaluated on two independent test sets including 85 and 50 WSIs, respectively, to show the robustness of the proposed method to differing staining protocols and scanner types. We also investigated the application of normalization as a pre-processing step by two techniques, the whole-slide image color standardizer (WSICS) algorithm, and a cycle-GAN based method. For the two independent datasets we obtained an AUC of 0.92 and 0.83 respectively. After rescanning the AUC improves to 0.91/0.88 and after style normalization to 0.98/0.97. In the future our algorithm could be used to automatically pre-screen prostate biopsies to alleviate the workload of pathologists.

Real-time deformable registration of multi-modal whole slides for digital pathology.

Digital pathology provides new ways to visualize tissue slides and enables new workflows for analyzing these slides. Analogous to radiology, adjacent tissue sections prepared with different stains or biomarkers (e.g. H&E, IHC, special stains, or ISH; chromogenic or fluorescent) may be seen as different modalities, each representing different structural and/or functional information. Today, the anatomic pathologist views multiple glass slides using an optical microscope and then combines the information in their head to reach a (diagnostic) opinion. Moreover, due to the nature of the slide preparation and digitization process, the tissue and its features do not have the exact same morphology, appearance, or spatial alignment, making it difficult to find the same region on adjacent slides. To address such concerns, this paper presents a method for the spatial alignment of multi-modal whole slide digital microscopy images. To remain practical, the described method employs a two-step registration strategy designed to reduce computation time: the first step computes a B-spline deformable transform on low-resolution images prior to visualization, the second step applies the precomputed transformation only to the high-resolution region currently being viewed. The proposed method is demonstrated using a number of cases comprising H&E and IHC stained slides. These results indicate the feasibility of deformable registration for spatial alignment of multi-modal whole slide digital microscopy images within practical time constraints.

Chemical modification of colloidal masks for nanolithography.

A method is presented to tune the holes in colloidal masks used for nanolithography. Using a simple wet-chemical method, a thin layer of silica is grown on masks of silica particles. The size of the holes is controlled by the amount of tetraethoxysilane (TEOS) added. More accurate tuning of the hole size is possible in the presence of a calibrated seed dispersion of silica colloids. We demonstrate modified masks that were used to create arrays of metal nanoparticles with a size ranging from 400 nm, for unmodified masks, down to tens of nanometers. The method is easy-to-use, fast, and inexpensive.

Combined optical tweezers/ion beam technique to tune colloidal masks for nanolithography.

A method is presented to control the in-plane ordering, size, and interparticle distance of nanoparticles fabricated by evaporation through a mask of colloidal particles. The use of optical tweezers combined with critical point drying gives single-particle position control over the colloidal particles in the mask. This extends the geometry of the colloidal masks from (self-organized) hexagonal to any desired symmetry and spacing. Control over the mask's hole size is achieved by MeV ion irradiation, which causes the colloids to expand in the in-plane direction, thus shrinking the size of the holes. After modification of the mask, evaporation at different angles with respect to the mask gives additional control over structure and interparticle distance, allowing nanoparticles of different materials to be deposited next to each other. We demonstrate large arrays of metal nanoparticles with dimensions in the 15-30 nm range, with control over the interparticle distance and in-plane ordering.

Colloidal epitaxy: playing with the boundary conditions of colloidal crystallization.

We have studied, with quantitative confocal microscopy, epitaxial colloidal crystal growth of particles interacting with an almost hard-sphere (HS) potential in a gravitational field and density matched colloids interacting with a long-range (LR) repulsive potential with a body-centred cubic (BCC) equilibrium crystal phase. We show that in both cases it is possible to grow thick, stacking fault-free metastable crystals: close-packed crystals with any stacking sequence, including hexagonal close packed (HCP), for the HS particles and face-centred cubic (FCC) in the case of the LR colloids. In accordance with recent computer simulations done for HS particles it was found that the optimal lattice constant to grow HS HCP crystals was larger than that of equilibrium FCC crystals. In addition, because of the absence of gravity, pre-freezing could be observed for the particles with the LR potential on a template of charged lines. We also argue that the ability to manipulate colloids with highly focused light, optical traps or tweezers, will become an important tool in both the study of colloidal crystallization and in making new structures. We show how cheap 2D and 3D templates can be made with optical tweezers and demonstrate, in proof of principle experiments with core-shell colloids, how light fields can generate crystal nuclei and other structures in the bulk of concentrated dispersions and how the effect of these structures on the rest of a dispersion can be studied quantitatively in 3D.