Virtum, treating cancer with an AI-powered image processing platform 

The challenge

Images so detailed that it is impossible to share 

 According to IBM data, up to 90% of modern healthcare data consists of images, including MRI scans, X-rays, histopathology images and many other types, depending on the stage and procedure. The more detailed the image is during the early diagnostic process, the more likely it is to spot early signs of a disease, when it may be easier to manage or even entirely prevent.  Later, these images are used to control the treatment process. Better quality may help identify suspicious or early changes in tissue or subtle changes in the patient’s condition, allowing medical professionals to initiate treatments earlier, ultimately leading to better recovery chances.  The bigger and more detailed the image, the more accurate the information about the patient is. The problem with this type of data is that it is challenging to process and share – a single image can be immense.  

Challenges with sharing make consultations complicated as healthcare professionals cannot simply send a scan to someone for consultation, even if it is necessary for a particularly complex diagnosis. Furthermore, medical tools are not AI-ready, fragmented, non-standardized and challenging to connect. By that, launching AI tools on medical imagery data poses an even greater challenge.  

Virtum aimed to provide a unified, standardized and convenient environment for working with medical imaging data. The goal of the creators was to deliver “Google Docs for medical images,” where one could share data with peers, store it, highlight specific parts and perform all the tasks known from modern, sophisticated office suites in a compliant and modern environment.  Moreover, it was also viewed as a way to annotate the data and to share a suspicious part with peers, to get a second or third opinion on the matter.   

The approach

Building a modern cloud office suite for healthcare imagery data 

Virtum is an outcome of a collaboration between histopathologists, scientists and software engineers. To make the tool as flexible and reliable as possible, the team decided to deliver a solution accelerator rather than the solution itself, which comes with multiple features that together enable the technology to tackle all the challenges:  

• Cloud-based technology for heavy images and computational workflows, scalable on demand and easily downscaled.  
• A pyramidal data viewer for smooth display and processing of huge images like histopathological scans or MRI images.  
• Cloud-based tool accessible from any computer and location.  
• Rich set of annotation tools with editing features, magic wand selection, and multiple brush sizes.  
• API integration with AI-powered tools for healthcare providers like data storage and annotation.  

The backend of the solution was developed entirely in Python using the Flask library and with support of MongoDB, which offers the necessary performance and flexibility. The frontend was built using the Qt platform, a framework designed for delivering sleek and multi-platform user interfaces.  The system is also AI-ready and accessible via API, allowing users to connect their own data with AI models to receive assistance with diagnosis.    

The outcome

Modern platform for better healthcare 

The deep understanding of the context, combined with technical proficiency, resulted in delivering a tool that has addressed the main challenges in the medical imaging and histopathology fields. 

Searchable and legible sample archive, quick access to any image

Contrary to more traditional methods of storing samples, these are fully searchable and can be tagged or marked in other ways to make them easy to find in the future. Over the course of many projects, Virtum had been adopted for work on High Performance Computing (HPC) clusters, for instance, Fortissimo. 

All-in-one workflow platform, tool-juggling no more 

Virtum provides the tools to store, manage, analyze and work on microscopic images. There is no need for the image to leave the secure environment of Virtum for any reason, bringing the level of security that hospitals and healthcare institutions require.  

Easy collaboration on any image, consult with peers from all around the world 

Virtum learns from the experiences of Google Workspace and similar collaboration tools, bringing the ease of sharing, commenting and editing to digital oncology and digital pathology systems. Users can check the version history, reply to comments in threads, share images and withdraw access at will.  

This presents an excellent opportunity for healthcare professionals to consult with peers from around the world on their cases. The user to whom the image was shared needs only a browser to access the full benefits of using Virtum and can support their colleagues in providing better care for patients.  

Cloud or on-prem infrastructure to better address the compliance

According to GDPR, it is often preferable for institutions to keep all data in-house. Technologies used to build Virtum support the on-prem infrastructure, leaving no uncertainty regarding compliance.  

AI-ready for both labelling and operations, boost your microscopic workflows with AI 

Virtum is accessible via API, making the tool ideal for Artificial Intelligence applications. Teams worldwide can utilize Virtum for data labeling to create an AI-ready dataset, training the algorithm to support their workflow.  

The algorithm can be further implemented within the Virtum workflow by allowing it to function as a separate component that connects with the rest of the suite via API. There is also no obstacle to using Virtum as a platform to run the AI algorithm and send the effects to other tools or workflows.  

Virtum is also powerful enough to combine artificial intelligence and radiology. Being a format-agnostic tool, it can deliver AI image diagnostics for radiology and serve as a platform for AI in pathology solutions. 

Originally, the tool was developed to support histopathologists. Adaptation of Virtum contributed to more efficient data modeling and helped enhance medical processes; however, it has also found applications in non-medical industries, including manufacturing, renewable energy, life sciences, biotech, space, mining and chemicals. The experiments involved microscope equipment manufacturers and medical R&D units to analyze images from remotely controlled microscopes, process super-resolution images, and take full advantage of the virtual microscopy toolbox.