Abstract:Electron cryo-tomography (cryo-ET) is a powerful imaging tool that allows three-dimensional visualization of subcellular architecture. During morphological analysis, reliable tomogram segmentation can only be achieved through high-quality data. However, unlike single-particle analysis or subtomogram averaging, the field lacks a useful quantitative measurement of cellular tomogram quality. Currently, the most prevalent method to determine cellular tomogram resolvability is an empirical judgment by experts, which is time-consuming. Methods like FSC between split tilt series suffer from severe geometrical artifacts. We address this gap with a neural network model to predict per-slice resolvability that can apply to tomograms collected from various species and magnifications. We introduce a novel metric, "TomoScore", providing a single-value evaluation of cellular tomogram quality, which is a powerful tool for pre-screening tomograms for subsequent automatic segmentation. We further explore the relationship between accumulated electron dose and resulting quality, suggesting an optimum dose range for cryo-ET data collection. Overall, our study streamlines data processing and reduces the need for human involvement during pre-selection for tomogram segmentation.

What problem does this paper attempt to address?

The main problem that this paper attempts to solve is the lack of effective quantitative quality assessment methods in cellular electron cryo - tomography (cryo - ET). Specifically: 1. **Background problems**: - Electron cryo - tomography (cryo - ET) is a powerful imaging technique that can achieve three - dimensional visualization of sub - cellular structures. - High - quality data is crucial for reliable tomogram segmentation and subsequent analysis. - Currently, the quality assessment of cellular tomograms mainly relies on the expert's empirical judgment, which is time - consuming and highly subjective. - Other methods such as FSC (Fourier Shell Correlation) between split - tilt series have serious geometric artifacts and are not suitable for the quality assessment of cellular tomograms. 2. **Research objectives**: - Develop a neural - network - based method that can automatically predict the resolvability of each slice and is applicable to cellular tomograms of different species and magnifications. - Introduce a new metric "TomoScore" to provide a single - value quality assessment of cellular tomograms, serving as a powerful tool for pre - screening tomograms for subsequent automatic segmentation. - Explore the relationship between the cumulative electron dose and image quality and suggest the optimal dose range for cryo - ET data collection. 3. **Solutions**: - Utilize the existing convolutional neural network design (ResNet101) to implement a generally applicable tomogram quality assessment tool - SliceQuality Network. - Train the model with the tomogram slice quality evaluated by humans to ensure that the model can accurately assess the quality of different tomograms. - Propose TomoScore, which normalizes the quality scores of all slices and then calculates the average to generate a single - value index that reflects the quality of the entire tomogram. - Verify the consistency and accuracy of the model by simulating tomograms with different cumulative doses and determine the optimal electron dose range. 4. **Expected results**: - Achieve automated, high - precision quality assessment of cellular tomograms, reduce human participation, and improve data processing efficiency. - Provide a standardized quality assessment tool for the cryo - ET community to promote the collection and analysis of high - quality data. In summary, this paper aims to solve the problem of the lack of effective quality assessment methods in cellular electron cryo - tomography by developing and applying neural network models, thereby improving the automation and accuracy of data processing.

TomoScore: A Neural Network Approach for Quality Assessment of Cellular cryo-ET

Isotropic reconstruction for electron tomography with deep learning

TomoNet: A streamlined cryoET software pipeline with automatic particle picking on flexible lattices

Automatic Localization and Identification of Mitochondria in Cellular Electron Cryo-Tomography Using Faster-Rcnn

A Complete Data Processing Workflow for Cryo-Et and Subtomogram Averaging

Tomo Live: an on-the-fly reconstruction pipeline to judge data quality for cryo-electron tomography workflows

Computational Methods for in Situ Structural Studies with Cryogenic Electron Tomography.

Training-free CryoET Tomogram Segmentation

AITom: Open-source AI platform for cryo-electron tomography data analysis

Attention-guided Quality Assessment for Automated Cryo-EM Grid Screening

ICON: 3D reconstruction with 'missing-information' restoration in biological electron tomography.

A survey of deep learning-based methods for cryo-electron tomography data analysis

CryoSamba: self-supervised deep volumetric denoising for cryo-electron tomography data

Simulating the cellular context in synthetic datasets for cryo-electron tomography

Rapid Synthesis of Cryo-ET Data for Training Deep Learning Models

SegET: Deep Neural Network with Rich Contextual Features for Cellular Structures Segmentation in Electron Tomography Image

Structure Detection in Three-Dimensional Cellular Cryoelectron Tomograms by Reconstructing Two-Dimensional Annotated Tilt Series

A method for restoring signals and revealing individual macromolecule states in cryo-ET, REST

Deep learning based supervised semantic segmentation of Electron Cryo-Subtomograms

Advances in cryo-electron tomography and subtomogram averaging and classification

A mammalian system for high-resolution imaging of intact cells by cryo-electron tomography