Hierarchical volumetric transformer with comprehensive attention for medical image segmentation.

Transformer is widely used in medical image segmentation tasks due to its powerful ability to model global dependencies. However, most of the existing transformer-based methods are two-dimensional networks, which are only suitable for processing two-dimensional slices and ignore the linguistic association between different slices of the original volume image blocks. To solve this problem, we propose a novel segmentation framework by deeply exploring the respective characteristic of convolution, comprehensive attention mechanism, and transformer, and assembling them hierarchically to fully exploit their complementary advantages. Specifically, we first propose a novel volumetric transformer block to help extract features serially in the encoder and restore the feature map resolution to the original level in parallel in the decoder. It can not only obtain the information of the plane, but also make full use of the correlation information between different slices. Then the local multi-channel attention block is proposed to adaptively enhance the effective features of the encoder branch at the channel level, while suppressing the invalid features. Finally, the global multi-scale attention block with deep supervision is introduced to adaptively extract valid information at different scale levels while filtering out useless information. Extensive experiments demonstrate that our proposed method achieves promising performance on multi-organ CT and cardiac MR image segmentation.

RGB Achromatic Metalens Doublet for Digital Imaging.

Chromatic aberration is a major challenge faced by metalenses. Current methods to achieve broadband achromatic operation in metalenses usually suffer from limited size, numerical aperture, and working bandwidth due to the finite group delay of meta-atoms, thus restricting the range of practical applications. Multiwavelength achromatic metalenses can overcome those limitations, making it possible to realize larger numerical aperture (NA) and sizes simultaneously. However, they usually require three layers, which increases their fabrication complexity, and have only been demonstrated in small sizes, with low numerical aperture and modest efficiencies. Here, we demonstrate a 1 mm diameter red-green-blue achromatic metalens doublet with a designed NA of 0.8 and successfully apply the metalens in a digital imaging system. This work shows the potential of the doublet metasurfaces, extending their applications to digital imaging systems such as digital projectors, virtual reality glasses, high resolution microscopies, etc.

High-throughput digital pathology via a handheld, multiplexed, and AI-powered ptychographic whole slide scanner.

The recent advent of whole slide imaging (WSI) systems has moved digital pathology closer to diagnostic applications and clinical practices. Integrating WSI with machine learning promises the growth of this field in upcoming years. Here we report the design and implementation of a handheld, colour-multiplexed, and AI-powered ptychographic whole slide scanner for digital pathology applications. This handheld scanner is built using low-cost and off-the-shelf components, including red, green, and blue laser diodes for sample illumination, a modified stage for programmable sample positioning, and a synchronized image sensor pair for data acquisition. We smear a monolayer of goat blood cells on the main sensor for high-resolution lensless coded ptychographic imaging. The synchronized secondary sensor acts as a non-contact encoder for precisely tracking the absolute object position for ptychographic reconstruction. For WSI, we introduce a new phase-contrast-based focus metric for post-acquisition autofocusing of both stained and unstained specimens. We show that the scanner can resolve the 388-nm linewidth on the resolution target and acquire gigapixel images with a 14 mm × 11 mm area in ∼70 seconds. The imaging performance is validated with regular stained pathology slides, unstained thyroid smears, and malaria-infected blood smears. The deep neural network developed in this study further enables high-throughput cytometric analysis using the recovered complex amplitude. The reported do-it-yourself scanner offers a portable solution to transform the high-end WSI system into one that can be made widely available at a low cost. The capability of high-throughput quantitative phase imaging may also find applications in rapid on-site evaluations.

Visual Sensor Image Analysis and Massage Techniques to Prevent and Treat Common Injuries of Sports Dance Practitioners.

In recent years, the performance of sports dance in China has become better and better. Naturally, the technical requirements for this dance are getting higher and higher, and the number and intensity of training have also increased, which has led to increasing injuries in sports dance. This article is based on visual sensor images to analyze and study the common injuries and prevention of sports dance practitioners. It is aimed at providing a certain reference basis for athletes' injuries, so that dance practitioners and coaches can better master sports dance training and teaching. Injury-related rules and prevention reduce the injury rate. This article puts forward the related technology of a visual sensor image and applies its technology to the prevention and research of common injuries in sports dance. At the same time, it analyzes the causes of sports dance practitioners' injuries and seeks economical and affordable massage techniques for prevention, and the method of treatment provides protection for dance practitioners. The experimental results in this article show that the Tuina group cured 15 subjects, 41 subjects were markedly effective, 13 subjects were improved, and 6 subjects were unhealed. The total effective rate was 92%.

A screening assistance system for cervical cytology of squamous cell atypia based on a two-step combined CNN algorithm with label smoothing.

BACKGROUND: Although many cervical cytology diagnostic support systems have been developed, it is challenging to classify overlapping cell clusters with a variety of patterns in the same way that humans do. In this study, we developed a fast and accurate system for the detection and classification of atypical cell clusters by using a two-step algorithm based on two different deep learning algorithms. METHODS: We created 919 cell images from liquid-based cervical cytological samples collected at Sapporo Medical University and annotated them based on the Bethesda system as a dataset for machine learning. Most of the images captured overlapping and crowded cells, and images were oversampled by digital processing. The detection system consists of two steps: (1) detection of atypical cells using You Only Look Once v4 (YOLOv4) and (2) classification of the detected cells using ResNeSt. A label smoothing algorithm was used for the dataset in the second classification step. This method annotates multiple correct classes from a single cell image with a smooth probability distribution. RESULTS: The first step, cell detection by YOLOv4, was able to detect all atypical cells above ASC-US without any observed false negatives. The detected cell images were then analyzed in the second step, cell classification by the ResNeSt algorithm, which exhibited average accuracy and F-measure values of 90.5% and 70.5%, respectively. The oversampling of the training image and label smoothing algorithm contributed to the improvement of the system's accuracy. CONCLUSION: This system combines two deep learning algorithms to enable accurate detection and classification of cell clusters based on the Bethesda system, which has been difficult to achieve in the past. We will conduct further research and development of this system as a platform for augmented reality microscopes for cytological diagnosis.

Extended range and aberration-free autofocusing via remote focusing and sequence-dependent learning.

Rapid autofocusing over long distances is critical for tracking 3D topological variations and sample motion in real time. Taking advantage of a deformable mirror and Shack-Hartmann wavefront sensor, remote focusing can permit fast axial scanning with simultaneous correction of system-induced aberrations. Here, we report an autofocusing technique that combines remote focusing with sequence-dependent learning via a bidirectional long short term memory network. A 120 µm autofocusing range was achieved in a compact reflectance confocal microscope both in air and in refractive-index-mismatched media, with similar performance under arbitrary-thickness liquid layers up to 1 mm. The technique was validated on sample types not used for network training, as well as for tracking of continuous axial motion. These results demonstrate that the proposed technique is suitable for real-time aberration-free autofocusing over a large axial range, and provides unique advantages for biomedical, holographic and other related applications.

GridFree: a python package of imageanalysis for interactive grain counting and measuring.

Grain characteristics, including kernel length, kernel width, and thousand kernel weight, are critical component traits for grain yield. Manual measurements and counting are expensive, forming the bottleneck for dissecting these traits' genetic architectures toward ultimate yield improvement. High-throughput phenotyping methods have been developed by analyzing images of kernels. However, segmenting kernels from the image background and noise artifacts or from other kernels positioned in close proximity remain as challenges. In this study, we developed a software package, named GridFree, to overcome these challenges. GridFree uses an unsupervised machine learning approach, K-Means, to segment kernels from the background by using principal component analysis on both raw image channels and their color indices. GridFree incorporates users' experiences as a dynamic criterion to set thresholds for a divide-and-combine strategy that effectively segments adjacent kernels. When adjacent multiple kernels are incorrectly segmented as a single object, they form an outlier on the distribution plot of kernel area, length, and width. GridFree uses the dynamic threshold settings for splitting and merging. In addition to counting, GridFree measures kernel length, width, and area with the option of scaling with a reference object. Evaluations against existing software programs demonstrated that GridFree had the smallest error on counting seeds for multiple crop species. GridFree was implemented in Python with a friendly graphical user interface to allow users to easily visualize the outcomes and make decisions, which ultimately eliminates time-consuming and repetitive manual labor. GridFree is freely available at the GridFree website (https://zzlab.net/GridFree).

Machine learning-based real-time object locator/evaluator for cryo-EM data collection.

In cryo-electron microscopy (cryo-EM) data collection, locating a target object is error-prone. Here, we present a machine learning-based approach with a real-time object locator named yoneoLocr using YOLO, a well-known object detection system. Implementation shows its effectiveness in rapidly and precisely locating carbon holes in single particle cryo-EM and in locating crystals and evaluating electron diffraction (ED) patterns in automated cryo-electron crystallography (cryo-EX) data collection. The proposed approach will advance high-throughput and accurate data collection of images and diffraction patterns with minimal human operation.

Automated processing of thermal imaging to detect COVID-19.

Rapid and sensitive screening tools for SARS-CoV-2 infection are essential to limit the spread of COVID-19 and to properly allocate national resources. Here, we developed a new point-of-care, non-contact thermal imaging tool to detect COVID-19, based on advanced image processing algorithms. We captured thermal images of the backs of individuals with and without COVID-19 using a portable thermal camera that connects directly to smartphones. Our novel image processing algorithms automatically extracted multiple texture and shape features of the thermal images and achieved an area under the curve (AUC) of 0.85 in COVID-19 detection with up to 92% sensitivity. Thermal imaging scores were inversely correlated with clinical variables associated with COVID-19 disease progression. In summary, we show, for the first time, that a hand-held thermal imaging device can be used to detect COVID-19. Non-invasive thermal imaging could be used to screen for COVID-19 in out-of-hospital settings, especially in low-income regions with limited imaging resources.

Computer-aided detection versus advanced imaging for detection of colorectal neoplasia: a systematic review and network meta-analysis.

BACKGROUND: Computer-aided detection (CADe) techniques based on artificial intelligence algorithms can assist endoscopists in detecting colorectal neoplasia. CADe has been associated with an increased adenoma detection rate, a key quality indicator, but the utility of CADe compared with existing advanced imaging techniques and distal attachment devices is unclear. METHODS: For this systematic review and network meta-analysis, we did a comprehensive search of PubMed/Medline, Embase, and Scopus databases from inception to Nov 30, 2020, for randomised controlled trials investigating the effectiveness of the following endoscopic techniques in detecting colorectal neoplasia: CADe, high definition (HD) white-light endoscopy, chromoendoscopy, or add-on devices (ie, systems that increase mucosal visualisation, such as full spectrum endoscopy [FUSE] or G-EYE balloon endoscopy). We collected data on adenoma detection rates, sessile serrated lesion detection rates, the proportion of large adenomas detected per colonoscopy, and withdrawal times. A frequentist framework, random-effects network meta-analysis was done to compare artificial intelligence with chromoendoscopy, increased mucosal visualisation systems, and HD white-light endoscopy (the control group). We estimated odds ratios (ORs) for the adenoma detection rate, sessile serrated lesion detection rate, and proportion of large adenomas detected per colonoscopy, and calculated mean differences for withdrawal time, with 95% CIs. Risk of bias and certainty of evidence were assessed with the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach. FINDINGS: 50 randomised controlled trials, comprising 34 445 participants, were included in our main analysis (six trials of CADe, 18 of chromoendoscopy, and 26 of increased mucosal visualisation systems). HD white-light endoscopy was the control technique in all 50 studies. Compared with the control technique, the adenoma detection rate was 7·4% higher with CADe (OR 1·78 [95% CI 1·44-2·18]), 4·4% higher with chromoendoscopy (1·22 [1·08-1·39]), and 4·1% higher with increased mucosal visualisation systems (1·16 [1·04-1·28]). CADe ranked as the superior technique for adenoma detection (with moderate confidence in hierarchical ranking); cross-comparisons of CADe with other imaging techniques showed a significant increase in the adenoma detection rate with CADe versus increased mucosal visualisation systems (OR 1·54 [95% CI 1·22-1·94]; low certainty of evidence) and with CADe versus chromoendoscopy (1·45 [1·14-1·85]; moderate certainty of evidence). When focusing on large adenomas (≥10 mm) there was a significant increase in the detection of large adenomas only with CADe (OR 1·69 [95% CI 1·10-2·60], moderate certainty of evidence) when compared to HD white-light endoscopy; CADe ranked as the superior strategy for detection of large adenomas. CADe also seemed to be the superior strategy for detection of sessile serrated lesions (with moderate confidence in hierarchical ranking), although no significant increase in the sessile serrated lesion detection rate was shown (OR 1·37 [95% CI 0·65-2·88]). No significant difference in withdrawal time was reported for CADe compared with the other techniques. INTERPRETATION: Based on the published literature, detection rates of colorectal neoplasia are higher with CADe than with other techniques such as chromoendoscopy or tools that increase mucosal visualisation, supporting wider incorporation of CADe strategies into community endoscopy services. FUNDING: None.

Comparative analysis of signal models for microscopic fractional anisotropy estimation using q-space trajectory encoding.

Microscopic diffusion anisotropy imaging using diffusion-weighted MRI and multidimensional diffusion encoding is a promising method for quantifying clinically and scientifically relevant microstructural properties of neural tissue. Several methods for estimating microscopic fractional anisotropy (µFA), a normalized measure of microscopic diffusion anisotropy, have been introduced but the differences between the methods have received little attention thus far. In this study, the accuracy and precision of µFA estimation using q-space trajectory encoding and different signal models were assessed using imaging experiments and simulations. Three healthy volunteers and a microfibre phantom were imaged with five non-zero b-values and gradient waveforms encoding linear and spherical b-tensors. Since the ground-truth µFA was unknown in the imaging experiments, Monte Carlo random walk simulations were performed using axon-mimicking fibres for which the ground truth was known. Furthermore, parameter bias due to time-dependent diffusion was quantified by repeating the simulations with tuned waveforms, which have similar power spectra, and with triple diffusion encoding, which, unlike q-space trajectory encoding, is not based on the assumption of time-independent diffusion. The truncated cumulant expansion of the powder-averaged signal, gamma-distributed diffusivities assumption, and q-space trajectory imaging, a generalization of the truncated cumulant expansion to individual signals, were used to estimate µFA. The gamma-distributed diffusivities assumption consistently resulted in greater µFA values than the second order cumulant expansion, 0.1 greater when averaged over the whole brain. In the simulations, the generalized cumulant expansion provided the most accurate estimates. Importantly, although time-dependent diffusion caused significant overestimation of µFA using all the studied methods, the simulations suggest that the resulting bias in µFA is less than 0.1 in human white matter.

A ball microscope for viewing the entire surface of amphibian embryos.

Waves on the surface of developing eggs/embryos need to be viewed from all sides of their 3D tissue. The ball microscope will enable tracking of cellular waves and determine their interactions with the cells on the surface. Nine microscopes are arrayed in a spherical formation around an imaging stage to create whole surface images of objects anywhere from 0.5 mm3 to 60 mm3 in size. The 3D printed ball-based microscope is made using nine, Opti-Tekscope OT-HD Digital USB Microscope Camera Magnifiers. Eight of the microscope cameras fit into the ball at 90° angles to each other and one bottom microscope is used for a base to hold the stage. The base will support a customised cuvette to hold the embryo in water. The microscopes are the size of a pen (13 cm long and 1 cm in diameter) and each have a ring light around their diameter for self illumination. The nine microscopes can be attached to a microcontroller for time-lapse automated imaging. This microscope will be compared to other microscopes developed for the same purpose. The microscope can be used for time lapse imaging of the surface of small 3D objects and can be used to view Axolotl salamander embryo development as the Axolotl embryos are 2 mm in diameter. Other amphibian eggs can also be imaged using this technique.

Comparison of traveling-subject and ComBat harmonization methods for assessing structural brain characteristics.

Multisite magnetic resonance imaging (MRI) is increasingly used in clinical research and development. Measurement biases-caused by site differences in scanner/image-acquisition protocols-negatively influence the reliability and reproducibility of image-analysis methods. Harmonization can reduce bias and improve the reproducibility of multisite datasets. Herein, a traveling-subject (TS) dataset including 56 T1-weighted MRI scans of 20 healthy participants in three different MRI procedures-20, 19, and 17 subjects in Procedures 1, 2, and 3, respectively-was considered to compare the reproducibility of TS-GLM, ComBat, and TS-ComBat harmonization methods. The minimum participant count required for harmonization was determined, and the Cohen's d between different MRI procedures was evaluated as a measurement-bias indicator. The measurement-bias reduction realized with different methods was evaluated by comparing test-retest scans for 20 healthy participants. Moreover, the minimum subject count for harmonization was determined by comparing test-retest datasets. The results revealed that TS-GLM and TS-ComBat reduced measurement bias by up to 85 and 81.3%, respectively. Meanwhile, ComBat showed a reduction of only 59.0%. At least 6 TSs were required to harmonize data obtained from different MRI scanners, complying with the imaging protocol predetermined for multisite investigations and operated with similar scan parameters. The results indicate that TS-based harmonization outperforms ComBat for measurement-bias reduction and is optimal for MRI data in well-prepared multisite investigations. One drawback is the small sample size used, potentially limiting the applicability of ComBat. Investigation on the number of subjects needed for a large-scale study is an interesting future problem.

Recent technical advances in whole slide imaging instrumentation.

Microscopic observation of biological specimen smears is the mainstay of diagnostic pathology, as defined by the Digital Pathology Association. Though automated systems for this are commercially available, their bulky size and high cost renders them unusable for remote areas. The research community is investing much effort towards building equivalent but portable, low-cost systems. An overview of such research is presented here, including a comparative analysis of recent reports. This paper also reviews recently reported systems for automated staining and smear formation, including microfluidic devices; and optical and computational automated microscopy systems including smartphone-based devices. Image pre-processing and analysis methods for automated diagnosis are also briefly discussed. It concludes with a set of foreseeable research directions that could lead to affordable, integrated and accurate whole slide imaging systems.

Diagnosis of some diseases such as cervical cancer is done using a microscope, and this process still relies heavily on human experts. Since the need for such diagnosis is increasing at a rapid pace, it makes a lot of sense to automate the whole process. This requires automatic microscopes, which should be able to take images of a 'slide' - a glass slab with colorized human cells at its surface. These images should get analyzed by a software, resulting in a fully automated diagnosis. This article reviews recent research into this field, especially the technical advances on the hardware for automated microscopes (also known as slide imagers). It compares research reports and highlights how there's still more effort needed to build low-cost, yet clinically useful systems. It also highlights some of the emerging technologies that can be integrated into slide imagers to enable new kinds of diagnostics.

PyPhase - a Python package for X-ray phase imaging.

X-ray propagation-based imaging techniques are well established at synchrotron radiation and laboratory sources. However, most reconstruction algorithms for such image modalities, also known as phase-retrieval algorithms, have been developed specifically for one instrument by and for experts, making the development and diffusion of such techniques difficult. Here, PyPhase, a free and open-source package for propagation-based near-field phase reconstructions, which is distributed under the CeCILL license, is presented. PyPhase implements some of the most popular phase-retrieval algorithms in a highly modular framework supporting its deployment on large-scale computing facilities. This makes the integration, the development of new phase-retrieval algorithms, and the deployment on different computing infrastructures straightforward. Its capabilities and simplicity are presented by application to data acquired at the synchrotron source MAX IV (Lund, Sweden).

Semi-automatic stitching of filamentous structures in image stacks from serial-section electron tomography.

We present a software-assisted workflow for the alignment and matching of filamentous structures across a three-dimensional (3D) stack of serial images. This is achieved by combining automatic methods, visual validation, and interactive correction. After the computation of an initial automatic matching, the user can continuously improve the result by interactively correcting landmarks or matches of filaments. Supported by a visual quality assessment of regions that have been already inspected, this allows a trade-off between quality and manual labour. The software tool was developed in an interdisciplinary collaboration between computer scientists and cell biologists to investigate cell division by quantitative 3D analysis of microtubules (MTs) in both mitotic and meiotic spindles. For this, each spindle is cut into a series of semi-thick physical sections, of which electron tomograms are acquired. The serial tomograms are then stitched and non-rigidly aligned to allow tracing and connecting of MTs across tomogram boundaries. In practice, automatic stitching alone provides only an incomplete solution, because large physical distortions and a low signal-to-noise ratio often cause experimental difficulties. To derive 3D models of spindles despite dealing with imperfect data related to sample preparation and subsequent data collection, semi-automatic validation and correction is required to remove stitching mistakes. However, due to the large number of MTs in spindles (up to 30k) and their resulting dense spatial arrangement, a naive inspection of each MT is too time-consuming. Furthermore, an interactive visualisation of the full image stack is hampered by the size of the data (up to 100 GB). Here, we present a specialised, interactive, semi-automatic solution that considers all requirements for large-scale stitching of filamentous structures in serial-section image stacks. To the best of our knowledge, it is the only currently available tool which is able to process data of the type and size presented here. The key to our solution is a careful design of the visualisation and interaction tools for each processing step to guarantee real-time response, and an optimised workflow that efficiently guides the user through datasets. The final solution presented here is the result of an iterative process with tight feedback loops between the involved computer scientists and cell biologists. LAY DESCRIPTION: Electron tomography of biological samples is used for a three-dimensional (3D) reconstruction of filamentous structures, such as microtubules (MTs) in mitotic and meiotic spindles. Large-scale electron tomography can be applied to increase the reconstructed volume for the visualisation of full spindles. For this, each spindle is cut into a series of semi-thick physical sections, from which electron tomograms are acquired. The serial tomograms are then stitched and non-rigidly aligned to allow tracing and connecting of MTs across tomogram boundaries. Previously, we presented fully automatic approaches for this 3D reconstruction pipeline. However, large volumes often suffer from imperfections (ie physical distortions) caused by the image acquisition process, making it difficult to apply fully automatic approaches for matching and stitching of numerous tomograms. Therefore, we developed an interactive, semi-automatic solution that considers all requirements for large-scale stitching of microtubules in image stacks of consecutive sections. We achieved this by combining automatic methods, visual validation and interactive error correction, thus allowing the user to continuously improve the result by interactively correcting landmarks or matches of filaments. We present large-scale reconstructions of spindles in which the automatic workflow failed and where different steps of manual corrections were needed. Our approach is also applicable to other biological samples showing 3D distributions of MTs in a number of different cellular contexts.

Accurate and practical feature extraction from noisy holograms.

Quantitative phase imaging using holographic microscopy is a powerful and non-invasive imaging method, ideal for studying cells and quantifying their features such as size, thickness, and dry mass. However, biological materials scatter little light, and the resulting low signal-to-noise ratio in holograms complicates any downstream feature extraction and hence applications. More specifically, unwrapping phase maps from noisy holograms often fails or requires extensive computational resources. We present a strategy for overcoming the noise limitation: rather than a traditional phase-unwrapping method, we extract the continuous phase values from holograms by using a phase-generation technique based on conditional generative adversarial networks employing a Pix2Pix architecture. We demonstrate that a network trained on random surfaces can accurately generate phase maps for test objects such as dumbbells, spheres, and biconcave discoids. Furthermore, we show that even a rapidly trained network can generate faithful phase maps when trained on related objects. We are able to accurately extract both morphological and quantitative features from the noisy phase maps of human leukemia (HL-60) cells, where traditional phase unwrapping algorithms fail. We conclude that deep learning can decouple noise from signal, expanding potential applications to real-world systems that may be noisy.

Digital Pathology Initiatives and Experience of a Large Academic Institution During the Coronavirus Disease 2019 (COVID-19) Pandemic.

CONTEXT.­: Pathology practices have begun integrating digital pathology tools into their routine workflow. During 2020, the coronavirus disease 2019 (COVID-19), caused by the severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), emerged as a pandemic, causing a global health crisis that significantly affected the world population in several areas, including medical practice, and pathology was no exception. OBJECTIVE.­: To summarize our experience in implementing digital pathology for remote primary diagnosis, education, and research during this pandemic. DESIGN.­: We surveyed our pathologists (all subspecialized) and trainees to gather information about their use of digital pathology tools before and during the pandemic. Quality assurance and slide distribution data were also examined. RESULTS.­: During the pandemic, the widespread use of digital tools in our institution allowed a smooth transition of most clinical and academic activities into remote with no major disruptions. The number of pathologists using whole slide imaging (WSI) for primary diagnosis increased from 20 (62.5%) to 29 (90.6%) of a total of 32 pathologists, excluding renal pathology and hematopathology, during the pandemic. Furthermore, the number of pathologists exclusively using whole slide imaging for primary diagnosis also increased from 2 (6.3%) to 5 (15.6%) during the pandemic. In 35 (100%) survey responses from attending pathologists, 21 (60%) reported using whole slide imaging for remote primary diagnosis following the Centers for Medicare and Medicaid Services waiver. Of these 21 pathologists, 18 (86%) responded that if allowed, they will continue using whole slide imaging for remote primary diagnosis after the pandemic. CONCLUSIONS.­: The pandemic served as a catalyst to pathologists adopting a digital workflow into their daily practice and realizing the logistic and technical advantages of such tools.

The 3D reconstructed skin micronucleus assay using imaging flow cytometry and deep learning: A proof-of-principle investigation.

The reconstructed skin micronucleus (RSMN) assay was developed in 2006, as an in vitro alternative for genotoxicity evaluation of dermally applied chemicals or products. In the years since, significant progress has been made in the optimization of the assay, including publication of a standard protocol and extensive validation. However, the diverse morphology of skin cells makes cell preparation and scoring of micronuclei (MN) tedious and subjective, thus requiring a high level of technical expertise for evaluation. This ultimately has a negative impact on throughput and the assay would benefit by the development of an automated method which could reduce scoring subjectivity while also improving the robustness of the assay by increasing the number of cells that can be scored. Imaging flow cytometry (IFC) with the ImageStream®X Mk II can capture high-resolution transmission and fluorescent imagery of cells in suspension. This proof-of-principle study describes protocol modifications that enable such automated measurement in 3D skin cells following exposure to mitomycin C and colchicine. IFC was then used for automated image capture and the Amnis® Artificial Intelligence (AAI) software permitted identification of binucleated (BN) cells with 91% precision. On average, three times as many BN cells from control samples were evaluated using IFC compared to the standard manual analysis. When IFC MNBN cells were visually scored from within the BN cell images, their frequency compared well with manual slide scoring, showing that IFC technology can be applied to the RSMN assay. This method enables faster time to result than microscope-based scoring and the initial studies presented here demonstrate its capability for the detection of statistically significant increases in MNBN frequencies. This work therefore demonstrates the feasibility of combining IFC and AAI to automate scoring for the RSMN assay and to improve its throughput and statistical robustness.

3DeeCellTracker, a deep learning-based pipeline for segmenting and tracking cells in 3D time lapse images.

Despite recent improvements in microscope technologies, segmenting and tracking cells in three-dimensional time-lapse images (3D + T images) to extract their dynamic positions and activities remains a considerable bottleneck in the field. We developed a deep learning-based software pipeline, 3DeeCellTracker, by integrating multiple existing and new techniques including deep learning for tracking. With only one volume of training data, one initial correction, and a few parameter changes, 3DeeCellTracker successfully segmented and tracked ~100 cells in both semi-immobilized and 'straightened' freely moving worm's brain, in a naturally beating zebrafish heart, and ~1000 cells in a 3D cultured tumor spheroid. While these datasets were imaged with highly divergent optical systems, our method tracked 90-100% of the cells in most cases, which is comparable or superior to previous results. These results suggest that 3DeeCellTracker could pave the way for revealing dynamic cell activities in image datasets that have been difficult to analyze.

Microscopes have been used to decrypt the tiny details of life since the 17th century. Now, the advent of 3D microscopy allows scientists to build up detailed pictures of living cells and tissues. In that effort, automation is becoming increasingly important so that scientists can analyze the resulting images and understand how bodies grow, heal and respond to changes such as drug therapies. In particular, algorithms can help to spot cells in the picture (called cell segmentation), and then to follow these cells over time across multiple images (known as cell tracking). However, performing these analyses on 3D images over a given period has been quite challenging. In addition, the algorithms that have already been created are often not user-friendly, and they can only be applied to a specific dataset gathered through a particular scientific method. As a response, Wen et al. developed a new program called 3DeeCellTracker, which runs on a desktop computer and uses a type of artificial intelligence known as deep learning to produce consistent results. Crucially, 3DeeCellTracker can be used to analyze various types of images taken using different types of cutting-edge microscope systems. And indeed, the algorithm was then harnessed to track the activity of nerve cells in moving microscopic worms, of beating heart cells in a young small fish, and of cancer cells grown in the lab. This versatile tool can now be used across biology, medical research and drug development to help monitor cell activities.