The image analysis software from MIPAR is a relatively new image analysis program. I have used ImageJ (Fiji) for about a decade now, but I find MIPAR to be much easier to use in terms of implementing and developing image analysis protocols and fully automated solutions.

MIPAR Analysis in Action

Below I highlight four examples of what you can achieve with the MIPAR image analysis software.

MIPAR can automatically identify, categorize, and count nuclei in immunohistological stains. Below is an image of an unidentified carcinoma immunohistologically stained for PAX8. The left half is the original image while the right half contains an overlay showing the different nuclei categorizations. A MIPAR image analysis protocol (called a recipe) was created to automatically identify the nuclei among the surrounding tissue and distinguish PAX8-positive from PAX8-negative nuclei. This recipe can be batch processed to automatically analyze hundreds of images and obtain nuclei counts for positive versus negative staining.

MIPAR can identify different regions of tissue sections after histological staining. Below is an image of a non-alcoholic fatty liver disease biopsy stained with Masson’s trichrome. The left half is the original image while the right half contains an overlay showing the different region categorizations. A MIPAR recipe was created to automatically identify fatty deposits (i.e., stenosis) and collagen deposits (fibrosis), and then determine the area fraction of each. This recipe can be batch processed, applied to hundreds of images, and used to automatically determine area fractions of distinctly-labeled regions.

MIPAR can identify and categorize fluorescence signals. The image below shows a differentiated PC12 cell fluorescently stained for nuclei (blue), microtubules (green), and mitochondria (red). The left half is the original image while the right half contains an overlay showing the different subcellular categorizations. A MIPAR recipe was created to identify mitochondria, to separate the cell body from the neurites, and then measure the number of mitochondria as an area fraction for each region. This recipe can be batched processed and automatically applied to hundreds of cells.

MIPAR can identify and categorize differences in DIC or phase images. The image below shows a wound healing assay where a portion of the cell monolayer has been scraped away. The left half is the original image while the right half contains an overlay showing the categorization of each region. A MIPAR recipe was created to automatically determine regions with and without cells and then measure the area of each. This recipe can be batch processed and automatically applied to hundreds of images, saving you countless hours in manual tracing (sigh, I wish I had this software during my postdoctoral research, where I measured cell migration manually!).

All of the recipes above, as well as many more pre-made recipes, are available for free to download from the MIPAR website (MIPAR recipe store). You can modify the pre-made recipes or make your own from scratch using MIPAR.

Deep Learning in MIPAR

There are some image analysis problems that do not yield to traditional image processing tools. This is where a deep learning algorithm may be the answer. Implementing the deep learning algorithm in MIPAR is quite easy. First, you trace the features of interest in a training image or images (or load your tracings created previously from other programs). Then you click a button to train the algorithm to identify your features of interest.

Once the deep learning training has been completed, you will have a deep learning model. This model can be applied to new images. When the model is run on an image, it outputs a probability map of a given pixel being part of your feature of interest. Once you have this map, you can use traditional tools to refine the image processing and measure your feature of interest.

The image above compares the best traditional image processing protocol with the MIPAR deep learning algorithm. A deep learning model was trained on 36 images that took 40 minutes to train. The deep learning algorithm can analyze a new image in 1.5 seconds, and batch process many images for a fraction of the time needed to manually trace them. Thus, the deep learning algorithm saves countless hours not spent manually tracing features.

Using MIPAR

Now that we have looked at what MIPAR can do, let’s take a look at the user interface for MIPAR.

In the image below, across the top are the 6 different applications in MIPAR. The Image Processor is where you can build traditional image analysis protocols (recipes) and run them on single images. The Batch Processor is where you can take your recipe and apply it to many images at once. The Real-Time Processor is similar to the Batch Processor, but it watches a specific folder and applies a specific recipe to files as they are added to the folder. The Post Processor is where you can review and clean-up the results of the Batch or Real-Time Processor. The 3D Toolbox is where you can visualize any Z-stack or time series data from the Batch or Real-Time Processors. The Deep Learning Trainer is where you can use MIPAR’s build-in deep learning algorithm to enhance your image analysis recipes.

In the image above, the Image Processor has been opened. One of the great things about the MIPAR software is that, unlike ImageJ, it performs non-destructive image analysis. This means that as image processing steps are applied to the image, the image is not permanently changed. Additionally, each step that has been applied can be seen in the recipe list (the far left box). If you are unhappy with the results of the current recipe, then you can flip back and forth between different steps and see how each affected your image. This is incredibly useful when developing your image analysis protocol. While developing my own protocols, I remember spending countless hours with ImageJ. I would try a protocol, find out at the end that it was insufficient, and then I would have to reload the image and start the protocol all over again.

See the MIPAR website for a full tour and demonstration of the software.

Learning MIPAR and Image Analysis

MIPAR has created a very useful and free course on Udemy: Learn Image Analysis. This course has two parts. The first part is an introduction to image analysis in general, which gives an overview of the steps of image analysis. In part two, each video explores a single concept in image analysis. They provide the images that are analyzed in each video so you can work along side the video. I highly recommend this course to learn the basics of image analysis and how to perform image analysis in MIPAR. They do not yet have any lessons on deep learning in MIPAR. However, they have a user manual and great customer support on the MIPAR website if you have any problems or questions.

CMIF Capabilities

We will soon have MIPAR image analysis software available on the CMIF workstations. However, you do not need to wait for that. As a faculty, staff, or student of Ohio State University, you will be able to get a license for free.

If you are part of the College of Engineering, you can request full licenses through the portal: https://ets.osu.edu/software.  If you are part of another college, you can email MIPAR at support@mipar.us and write how many licenses you would like for yourself or your lab.