Artificial Intelligence, Machine Learning & Deep Learning

Boost your research with advanced AI methods

With Machine Learning and Deep Learning methods in arivis Vision4D segmentation of multi-channel images becomes a quick and easy task. No need to be an expert in bioinformatics: with your scientific expertise you mark and classify structures of interest in your samples and let the cloud train your neural network. Your benefit: you get reliable results in hours rather than weeks or even months of extensive manual work.

Easy to use, guided workflow specifically for users with little knowledge in image analysis
Definition of expected results by labelling a few lines in the image
Creation of robust and reliable analysis results
Less time-to-results by re-using your training data on various data sets
Fully integrated workflows with support for open platforms and various standards
Direct and fast feedback with previews of results and probability maps
Much faster and more reproducible compared to manual segmentation
Also for experts worth a try when other algorithms fail or are very tedious

deep-learning-infographic-vision4d-arivis-web

Talk with an Expert

Why use Machine Learning

A conventional algorithm is designed by a specialist to answer exactly one question. In contrast, a Machine Learning algorithm can be adapted to a wide variety of questions, simply by training it. The Machine Learning algorithm “learns” patterns and adapts itself.

Machine Learning allows to use this expert knowledge of the user by drawing and classifying some samples of structures of interest into the image. The subsequent automatic training uses this information to automatically create the algorithm to be used to find these structures all over this and other images.

APPLICABLE Applicable to images of any number of dimension and unlimited size

EASY TO USE Easy to use and guided workflow, no coding required

FAST FEEDBACK Fast feedback during training with previews of results

EASILY SCALABLE Batch analysis compatible to easily scale your image analysis process

Machine Learning Workflow

Our new UI and the smooth workflow integration guides the customer in a few steps through the process:

TRAIN Use specific user knowledge to recognize and classify structures of interest

SAVE Save or further improve your training by adding more labels any time

APPLY Start segmentation with a single click or use the Batch Analysis Module to process hundreds of data sets at once

Deep Learning Workflow

deep_learning_workflow

Acquire Imaging Data Vision4D supports a wide variety of multidimensional imaging modalities and file formats, thanks to the unique arivis SIS Converter. Never worry about compatibility again!

Train Your Model Perform initial training of your specific neuronal network with the APEER cloud platform, ZEISS ZEN Intellesis, or various well-established scientific open source AI training platforms

Deep Learning Inference Import your neuronal network directly into Vision4D and let it do the analysis work on your imaging datasets. Typical tasks are reduced from weeks to hours compared to manual execution.

Vision4D Video Tutorials

How to use Deep Learning

Deep Learning Workflow with APEER and ZEN Intellesis

This tutorial demonstrates the workflow for deep learning applications with APEER, ZEN Intellesis and arivis Vision4D. Vision4D 3.6 fully supports import of deep learning models in the Open Neural Network Exchange (ONNX) format as well as the CZ model from ZEISS ZEN Intellesis and the APEER cloud-based image processing platform. With APEER you easily annotate and train new models or combine already existing ones into customized workflows, without the need for programming knowledge. Vision4D now gives you the flexibility to import and apply fully trained neural networks from various libraries and backends to your analysis pipelines to enable deep learning for automated analysis of your imaging data.

Create annotations for Deep Learning and apply automated AI segmentation

Hippocampus tissue section, transmission electron microscopy. In total, 30 TEM serial sections were used with 309 mitochondria objects, annotated manually with Vision4D 3.6. This manual annotation was used to train an ONNX U-Net Deep Learning model that was then applied on the whole dataset in Vision4D 3.6 for automated segmentation. In a following step, we calculated the ratio of the mean intensity of each object to its volume. This radiometric function reflects the differences in the mitochondria phenotype. These differences are visualized as a color-coded classification of the objects. Original imaging data was kindly provided by Dr. Wendy Bautista, MD PhD, Barrow Neurological Institute, Phoenix Children’s Hospital.

Full integration in arivis Vision4D

Machine Learning in arivis Vision4D is a fully integrated solution which allows to combine the segmentation results with any other functionality of the pipeline.

These include:

Filtering (e.g. on volumes, intensities, etc.)
Tracking
Grouping
Distance measurements
In addition, you can also use the probability map as a basis for a subsequent segmentation in the arivis Analysis Pipeline
Deep Learning supports an integrated workflow with various options for connected software and training platforms

arivis Machine Learning works for several different multi-dimensional images from many modalities in microscopy:

Fluorescence Microscopy
Superresolution Microscopy
Transmission Light or Label Free Microscopy
Confocal Microscopy
Light Sheet Microscopy
Electron Microscopy (2D and 3D)
X-ray Microscopy
CT and MRT images

All image formats supported by arivis Vision4D will also be compatible with Machine Learning. CZI, TIFF, JPG, PNG, TXM and all Bio-Formats compatible images.

The arivis Imaging Platform

The arivis Imaging Platform is a flexible computing universe for Imaging Science that scales, parallelizes, integrates and connects all imaging workflows, sparking organization-wide image data proficiency and efficiency at all levels. The integrated toolsets take care of everything from the file storage format to user and project-specific computations to reporting. The computational and management hubs that comprise the platform connect your datasets and take care of your central imaging databases and can expose data assets - including raw data and specified portions of raw data - to Machine Learning and AI routines.

VisionHub Your nucleus of computational activity and collaboration at scale. VisionHub enables design and execution of large-scale experiments that produce results from images, whether datasets are already stored or are actively being produced. Mass quantities of imaging datasets are onboarded and processed by a virtual team of computational workers. Smartly utilize processing cores in servers or workstations with the maximum speeds possible. Read More

Vision4D

Your modular software platform for extracting results from scientific images. Adjusts to your hardware for maximum power and smooth interactivity even with large image datasets of virtually unlimited size. Bring diverse tools into one environment and enable your users to connect them into productive workflows. Highly interactive for optimization and quality at every step. Extend Vision4D directly via Python and connect it to other applications via libraries.

VisionVR Display real image data in Virtual Reality by utilizing patent-pending direct volume rendering techniques with no need to convert data or create surface models. Directly use your hands to move, rotate, scale, and shape your digital image data. Interactively proofread, edit, track or segment multi-dimensional images from nearly any source instrument. Discuss your findings and results worldwide in a collaborative connected environment. Read More

OUR TEAM IS READY TO HELP

Use the form below to get in contact with our arivis experts. Please describe your query and our team will contact you.