Image based cellular contractile force evaluation with small-world network inspired CNN: SW-UNet

Diffusion model predicts the geometry of actin cytoskeleton from cell morphology

Cells exhibit various morphological characteristics due to their physiological activities, and changes in cell morphology are inherently accompanied by the assembly and disassembly of the actin cytoskeleton. Stress fibers are a prominent component of the actin-based intracellular structure and are highly involved in numerous physiological processes, e.g., mechanotransduction and maintenance of cell morphology. Although it is widely accepted that variations in cell morphology interact with the distribution and localization of stress fibers, it remains unclear if there are underlying geometric principles between the cell morphology and actin cytoskeleton. Here, we present a machine learning system that uses the diffusion model to convert the cell shape to the distribution and alignment of stress fibers. By training with corresponding cell shape and stress fibers datasets, our system learns the conversion to generate the stress fiber images from its corresponding cell shape. The predicted stress fiber distribution agrees well with the experimental data. With this conversion relation, our system allows for performing virtual experiments that provide a visual map showing the probability of stress fiber distribution from the virtual cell shape. Our system potentially provides a powerful approach to seek further hidden geometric principles regarding how the configuration of subcellular structures is determined by the boundary of the cell structure; for example, we found that the stress fibers of cells with small aspect ratios tend to localize at the cell edge while cells with large aspect ratios have homogenous distributions.

LSW-Net: Lightweight Deep Neural Network Based on Small-World properties for Spine MR Image Segmentation

BACKGROUND

Segmenting spinal tissues from MR images is important for automatic image analysis. Deep neural network-based segmentation methods are efficient, yet have high computational costs.

PURPOSE

To design a lightweight model based on small-world properties (LSW-Net) to segment spinal MR images, suitable for low-computing-power embedded devices.

STUDY TYPE

Retrospective.

POPULATION

A total of 386 subjects (2948 images) from two independent sources. Dataset I: 214 subjects/779 images, all for disk degeneration screening, 147 had disk degeneration, 52 had herniated disc. Dataset II: 172 subjects/2169 images, 142 patients with vertebral degeneration, 163 patients with disc degeneration. 70% images in each dataset for training, 20% for validation, and 10% for testing.

FIELD STRENGTH/SEQUENCE

T1- and T2-weighted turbo spin echo sequences at 3 T.

ASSESSMENT

Segmentation performance of LSW-Net was compared with four mainstream (including U-net and U-net++) and five lightweight models using five radiologists' manual segmentations (vertebrae, disks, spinal fluid) as reference standard. LSW-Net was also deployed on NVIDIA Jetson nano to compare the pixels number in segmented vertebrae and disks.

STATISTICAL TESTS

All models were evaluated with accuracy, precision, Dice similarity coefficient (DSC), and area under the receiver operating characteristic (AUC). Pixel numbers segmented by LSW-Net on the embedded device were compared with manual segmentation using paired t-tests, with P < 0.05 indicating significance.

RESULTS

LSW-Net had 98.5% fewer parameters than U-net but achieved similar accuracy in both datasets (dataset I: DSC 0.84 vs. 0.87, AUC 0.92 vs. 0.94; dataset II: DSC 0.82 vs. 0.82, AUC 0.88 vs. 0.88). LSW-Net showed no significant differences in pixel numbers for vertebrae (dataset I: 5893.49 vs. 5752.61, P = 0.21; dataset II: 5073.42 vs. 5137.12, P = 0.56) and disks (dataset I: 1513.07 vs. 1535.69, P = 0.42; dataset II: 1049.74 vs. 1087.88, P = 0.24) segmentation on an embedded device compared to manual segmentation.

DATA CONCLUSION

Proposed LSW-Net achieves high accuracy with fewer parameters than U-net and can be deployed on embedded device, facilitating wider application.

EVIDENCE LEVEL

2.

TECHNICAL EFFICACY

1.

Profiling oocytes with neural networks from images and mechanical data

The success rate of assisted reproductive technologies could be greatly improved by selectively choosing egg cells (oocytes) with the greatest chance of fertilization. The goal of mechanical profiling is, thus, to improve predictive oocyte selection by isolating the mechanical properties of oocytes and correlating them to their reproductive potential. The restrictions on experimental platforms, however - including minimal invasiveness and practicality in laboratory implementation - greatly limits the data that can be acquired from a single oocyte. In this study, we perform indentation studies on human oocytes and characterize the mechanical properties of the zona pellucida, the outer layer of the oocyte. We obtain excellent fitting with our physical model when indenting with a flat surface and clearly illustrate localized shear-thinning behavior of the zona pellucida, which has not been previously reported. We conclude by outlining a promising methodology for isolating the mechanical properties of the cytoplasm using neural networks and optical images taken during indentation.

Wrinkle force microscopy: a machine learning based approach to predict cell mechanics from images

Combining experiments with artificial intelligence algorithms, we propose a machine learning based approach called wrinkle force microscopy (WFM) to extract the cellular force distributions from the microscope images. The full process can be divided into three steps. First, we culture the cells on a special substrate allowing to measure both the cellular traction force on the substrate and the corresponding substrate wrinkles simultaneously. The cellular forces are obtained using the traction force microscopy (TFM), at the same time that cell-generated contractile forces wrinkle their underlying substrate. Second, the wrinkle positions are extracted from the microscope images. Third, we train the machine learning system with GAN (generative adversarial network) by using sets of corresponding two images, the traction field and the input images (raw microscope images or extracted wrinkle images), as the training data. The network understands the way to convert the input images of the substrate wrinkles to the traction distribution from the training. After sufficient training, the network is utilized to predict the cellular forces just from the input images. Our system provides a powerful tool to evaluate the cellular forces efficiently because the forces can be predicted just by observing the cells under the microscope, which is much simpler method compared to the TFM experiment. Additionally, the machine learning based approach presented here has the profound potential for being applied to diverse cellular assays for studying mechanobiology of cells.

A deep learning approach, termed wrinkle force microscopy, allows for conducting traction force microscopy by observing bright-field cell images but without using its conventional requirements such as fluorescent microbeads and confocal microscopy.