AI software: "iSTIX opens your world to the possibilities of digital pathology"

The healthcare market offers a multitude of microscopes that make cells visible to the human eye. The same applies to AI-based software for image analysis. After taking the microscopic images, scientist are faced with large volumes of scans with usually low resolution. Yet when all aspects merge together, they open up a the world of digital pathology.

Volker Bruns sat down for an interview with MEDICA.de and talked about the integration of Fraunhofer’s AI-based iSTIX software into a microscope-centered workflow. He also details the software’s medical applications and reveals why this technology shows great promise for educational and precision medicine purposes.

Mr. Bruns, how does the Fraunhofer iSTIX software complement the microscope procedure?

Volker Bruns: The standard software that is included with a microscope camera only allows you to view and record a specific image section. It is actually even slightly smaller than the section you can see through the eyepiece.

Meanwhile, iSTIX allows for conventional microscopes to be retrofitted to turn into a whole-slide scanner. iSTIX analyzes the microscope camera’s video feed in real-time and fuses the individual image tiles together into a whole slide image (WSI). The advantage of a WSI is that you can obtain an overview of all images at low magnification, while it also allows you to subsequently identify details at larger or original viewing magnification. The slides can then be exported to multiple formats or uploaded directly into cloud storage to use them for educational purposes or to request a second opinion for example.

The slide scanning process opens your world to the possibilities of digital pathology. It also allows you to automate repetitive tasks like counting cells. Deep learning technologies can handle even the most complex analytical processes. This could be an important building block in managing the increasing number of assessments associated with an aging population and facilitating better options as it relates to precision medicine in the future.

What are aspects to consider when you integrate the software into a microscope-centered workflow?

Bruns: The camera is a crucial component. We recommend using a 5-megapixel color image camera that can deliver at least 20 frames per second. In addition, you also need a so-called global shutter to ensure great results, because the images are otherwise somewhat distorted and can no longer be stitched seamlessly. That being said, there are many cameras on the market that meet these criteria. Motion unsharpness is caused by rapid movement of the stage - to prevent this, you can increase the microscope’s light intensity to decrease camera exposure. Since the background illumination isn’t always homogenous, iSTIX includes an easy-to-calibrate backlight correction feature: you only need to hit 1-3 places in the background one time to subsequently use them as reference images.

What are the medical applications of the iSTIX software?

Bruns: iSTIX was specifically designed for life science applications using transmitted-light paths such as brightfield microscopy, including histology, hematology, cytology or microbiology. The samples just need to have enough detail as image registration is impossible with two mostly blank images. As a cost-effective and affordable solution, iSTIX is particularly suited for educational purposes. You can scan interesting specimens and upload them to a web platform.

What makes image registration technology - the heart of the software - so unique?

Bruns: The registration algorithm is fast enough to accommodate the video stream in real-time even on laptops. To reduce memory usage and computing time, the number of individual image tiles that are being added to the panorama is intelligently minimized. At the same time, all live images are visualized to ensure smooth navigation. If a new image is sharper than the previous one, the latter will be replaced. To make sure that memory usage during image recording does not grow exponentially, each tile is likewise reduced in real-time in several stages. Depending on the current view zoom range, the best resolution can be used right away. However, should the algorithm lose the connection for some reason - for example, if the user moves the stage too abruptly-, you do not have to revert back to the exact location. You only have to return to a point in the existing panorama.

We are working on further innovations such as finding a way to register different magnifications to each other or facilitating automatic lens change detection.

What is the advantage over a standard digital slide scanner?

Bruns: The main advantage is that iSTIX costs less and is accessible at your own desk at any times. For example, if I want to generate a Ki-67 scoring protocol on a few samples, I can quickly scan the relevant areas and transfer them to the Cognition Master image analysis software made by our partner VMscope with a click - without the need to get up or block the scanner.

Another advantage is that it also allows stronger magnifications that require oil immersion. Not many scanners are currently able to support this function. Having said that, as a manual scanning solution, iSTIX is unable to keep up with the throughput of a fully automated scanner.

You will give a lecture about the Fraunhofer iSTIX software at the MEDICA LABMED FORUM 2019. What does this mean to you?

Bruns: I am honored to be invited to showcase and represent our experiences and views on digital pathology. As Fraunhofer associates, we look through an academic lens and appreciate the exponential growth of technological advancements, especially due to artificial intelligence. However, as we direct our focus on the needs of the industry, we are also aware that not very many innovations have made the leap into actual work environments. The healthcare market places high demands on developers. I look forward to discussing this with the other Forum participants.