AI Segmentation

These modules are included with sliceOmatic. The AI modules will let you analyze slices one by one or as 3D volumes.

The AI modules enable you to train your own AI with your data. The AI modules can be used to segment images with pre-defined trained AI weights. If none of the available pre-trained weights match your requirements, you can train your own AI with your own data directly in these module. Everything you need to train an AI is provided. Just follow a few simple configuration steps, start the training, wait for it to finish (that step may need some patience...) save the resulting weights and use the newly trained AI to segment your images. You can use this process iteratively. First do some images manually or semi-automatically with the standard sliceO's interface, then train an AI with these few images. Use the trained AI as a first step in your segmentation process, correcting any errors with the sliceO many segmentation modules. Then when you have more images, re-train the AI. Eventually your AI will be good enough to do the segmentation of your images automatically. If you are familiar with AI technologies and can code in Python, these modules let you costumize the AI models and metrics as you please.

The ABACS module, is a plugin created by Voronoi Health Analytics.

The ABACS module does an automatic segmentation of CT slices at the T4 or L3 level. With the simple click of a button, the ABACS (Automated Body composition Analyzer using Computed tomography image Segmentation) module will identify and tag the muscle, sub-cutaneous, visceral and intra abdominal fat on all the L3 or T4 slices you have selected, without any further user intervention.

Features

The process of identifying each tissue inside an image is known as "segmentation". And this is where sliceOmatic outshines the competition: its segmentation tools.

sliceOmatic balances powerful segmentation tools with a highly interactive interface, enabling you to completely control the segmentation process.

Different modalities or even different tissues within the same slice are best segmented with different techniques. SliceOmatic makes it easy for you to use the appropriate technique for each specific task. For example, in an MRI slice of the abdomen, you may want to use mathematical morphology to segment the sub-cutaneous fat and region-growing to segment the intra abdominal fat.

sliceOmatic is a research tool:

As a researcher, analysing the images is only a means to an end. What you really want is getting accurate anatomical volumes. SliceOmatic is designed as a research tool. In addition to helping you create the great images for these "wow" slides in your presentations, it will export the surfaces and volumes of each segmented tissue in a spread-sheet compatible file.

Maybe your area of research is biomechanics? Then you will want to use sliceOmatic to create polygonal meshs and/or countour lines from the segmented data. These can then be exported in a number of standard formats to be used either in finite element analisys or to create stereo-lithographic models.

The wide range of formats that sliceOmatic supports makes it an ideal tool to both view and analyze all your images. You can see the list of supported image formats .

"I wanted to report back to you the outstanding results obtained with the new slice0matic software. The system works beautifully and the new software greatly eases analysis of MRI scans. Please extend my thanks to Yves for a job well done."

Dr. Steve Heymsfield - St Luke Roosevelt Hospital, NY.

Speed

The AI segmentation modules let you analyze slices in as little as 20msec per slice!

In a technical publication, Dr Bonekamp and his colleagues compared sliceOmatic with 4 other image analysis packages. Not only did they found that sliceOmatic was among the easiest to use and had the best graphical user interface, but they also found that segmenting adipose and visceral fat with sliceOmatic was at least 3 time faster than with any of the other software.

"[...] We consider sliceOmatic a very good choice for segmentation of abdominal adipose tissue on MR images. [...]"

S Bonekamp, P Ghosh, S Crawford et all, Quantitative comparison and evaluation of software packages for assessment of abdominal adipose tissue distribution by magnetic resonance imaging , Int J of Obesity, 32:100-111, 2008

Additional Modules

Some of the Additional Modules available:

• Tool: Automation
• Tool: CT Gradient
• Tool: DICOM Compare
• Tool: DICOM Info
• Tool: Pixel Interpolation
• Class: Border
• Class: Clone
• Class: Gradient Amplification
• Class: RR2D
• Class: RR3D
• Mode: Histogram Segmentation
• Mode: TAG Interpolation
• Mode: MR Spectroscopy

Tool: Automation

This tool enable you to automate some actions.  It can be used to read files and perform scripts for each files, or perform scripts for each frames, or simply perform scripts a number of times.  For example it can be used to take one snapshot of each frames, or to create an animation of a 3D object rotating, or in an industrial setting, automate the analysis of images acquire from a continuous process.

For more information, please consult the sliceOmatic 5.0 Online User's Manual

Tool: CT Gradient

This tool is used to limit the gradient computation to a threshold range.

Some of the segmentation modes (Morpho Á Snakes) are based on the gradient of the pixel values. For CT images, we can also limit the gradient computation to a desired threshold, this give us a better segmentation.

For more information, please consult the sliceOmatic 5.0 Online User's Manual

Tool: DICOM Compare

This tool integrate the standalone tool DICOM_Compare from TomoVision inside sliceOmatic. It enable you to see the DICOM tags associated with any of the loaded frames. You can either compare these tags with the DICOM definition for the same modality or with the tags of another frame. This is very useful if you want to use the Mixer, Multiplex or DICOM Tree classes.

For more information, please consult the sliceOmatic 5.0 Online User's Manual

Tool: DICOM Info

This tool is designed to work in conjuction with the "2D Results" tool.  It enable to save up to 4 DICOM tag values in the Excel result file.

The 4 tags that are defined in the interface of the "DICOM Info" tool will be seen by the configuration page of the "2D Results" tool and can be selected. If some of the 4 tags are selected for the result file, the value of these DICOM tags, for each selected slice, will be saved in the Excel result file.

For more information, please consult the sliceOmatic 5.0 Online User's Manual

Tool: Pixel Interpolation

This module was developed by TomoVision for Yajima Nobuhiko of the Image Labo in Japan.

This tool enable you to "smooth" the images at higher magnifications.

For more information, please consult the sliceOmatic 5.0 Online User's Manual

Class: Border

This class will enable you to add borders around the frames.

This can be useful if part of the information you want to segment is outside the field of view. Adding a border enable you to extrapolate outside the frame.

For more information, please consult the sliceOmatic 5.0 Online User's Manual

Class: Clone

This module was developed by TomoVision for Yajima Nobuhiko of the Image Labo in Japan.

This class will create 2 branches in the database tree. One branch contain the original images, the other an exact clone of these images. Different treatment can be applied on the GLI images of each branch, but the segmentation TAG images will be identical in both the original and the cloned branches.

This class can be useful to compare the results of other classes that modify the GLI images (such as "Filter" or "Gradient"). For example, if you create a tree with the classes: Root / Clone / Filter / Frame, you can then apply different filters to the original and clone branches and compare the results. With that tree, the shortcut key "Insert" / "Delete" will toggle between the original and cloned images.

For more information, please consult the sliceOmatic 5.0 Online User's Manual

Class: Gradient Amplification

This module was developed by TomoVision for the LIO group of the ETS in Montreal, Canada.

This class enable you to artificially increase the gradient around a specific target GLI value. This will help segmentation techniques that are gradient based (such as "Morpho" and "Snake"). It can be used for example in segmenting CT images where you know you want to segment at specific Hounsfield values but do not want to use threshold based techniques. By increasing the gradient at the desired Hounsfield value, you insure that the gradient based technique will find the correct boundary.

For more information, please consult the sliceOmatic 5.0 Online User's Manual

Class: RR2D (Rigid Registration 2D)

This class will enable you to register slices in 2D (in the plane of the slices).  It can be used, for example, to re-align series if you need to move a patient between acquisitions.  This is a landmark based registration technique.

For more information, please consult the sliceOmatic 5.0 Online User's Manual

Class: RR3D (Rigid Registration 3D)

This class will enable you to register series in 3D using rigid transformations (translation, rotation and scaling).  This is a landmark based registration technique.

For more information, please consult the sliceOmatic 5.0 Online User's Manual

Mode Histogram Segmentation

This module was developed by TomoVision for Jeffrey Cooley of McMurdoch University in Australia.

This mode is used to segment the images with the help of the histogram.

For more information, please consult the sliceOmatic 5.0 Online User's Manual

Mode TAG Interpolation

In this mode, you create Tag values on frames that do not have them by interpolating between frames that do.

This mode is useful for 3D structures that do not change much between slices (ex: the long bones of the leg).  You only need to segment a few slices and then let this mode interpolate the missing slices from the data on the existing slices.

For more information, please consult the sliceOmatic 5.0 Online User's Manual

Mode MR Spectroscopy

This module was developed by TomoVision for Robert Ross of Queens University in Kingston, Canada.

This mode enable you to analyse MR spectroscopy data.

Note:
At this time sliceO only recognise Siemens datasets. If you have any other kind of datasets, please contact us! We need sample data...

For more information, please consult the sliceOmatic 5.0 Online User's Manual

Security

I do understand that many hospitals have worries before installing a software that can access patient data on their system.

So, here's a few information on sliceOmatic that, I hope, will answer the questions you may have.

Data sharing: SliceOmatic does not communicate with any other devices, database, the cloud or TomoVision. It does not store any information outside itself. It does not even talk to the PACS! (All the images it analyze must be first copied from the PACS to the local PC).

The only patient information it access is the one present in the original DICOM files, and these are not copied or kept in any database. When you close the application, no data is retain anywhere.

Exported files: The program create 3 types of files derived from the original images:

- Segmentation file (.tag) that will have a map of the segmentation. These files only contain pixel information and no patient data. By default, these files are saved in the same location as the original DICOM files. So, either locally on your PC or on a shared drive accessible by Windows.

- Result file (.csv) that contain the surfaces/volumes computed from the original files. These file can contain patient information if these info have been selected for saving. The content and location of these results files are under the user's control.

- AI "Weight" files. If, to automate part of the segmentation process, the user train an AI on some images the program can save the computed AI weights. These files do not contain any patient information. They are saved locally and are not shared with TomoVision.

Anonymous mode: If you are doing a "blind" study, sliceOmatic has an "Anonymous" mode that can be enabled to prevent the user from seeing any patient information or saving it to the results files.