System for annotating X-ray images and generating pathologies

Type Web application
TECHNOLOGIES Node.JS, OpenCV, Python, Typescript
TEAM 1 ML developer, 1 Web-developer, 1 Frontend developer, 1 Back-end developer, 1 Analyst, 1 UX/UI Designer, 1 Project manager

Creating medical software is challenging due to strict regulations and formalization in the healthcare sector. Many annotation tools exist for professionals, including a system for annotating X-ray images, but some don’t comply with laws, like storing patient data in specific locations, and may not fulfill certain functions. This can make the software less user-friendly for various professionals, leading to inefficient use of time spent learning the interface. 

The project’s main goal was the development of a cloud-based web platform with multi-user access, enabling healthcare professionals to work with X-ray studies within a unified environment, including original, annotated, and artificially generated studies.

What we did

  1. Cloud-based web platform (system for annotating X-ray images) with multi-user access
  2. Toolset for X-ray image annotation
  3. Capability to upload, store, and download data
  4. Functionality for collaborative user work
  5. Pathology generator, which uses neural networks to create pathologies on an image of a healthy lung based on specified parameters.
  1. Designing a user-friendly interface;
  2. Convenient and Intuitive Annotation Tool for Radiologists;
  3. Development of the Pathology Generator.
  1. We identified and introduced essential data about the studies that might be needed for comfortable work with them. These include the study's name, status, study type, creation date, update date, and a list of actions that can be performed on the study.We recognized the user's need to identify the current stage of the study. Consequently, we introduced study statuses that users can assign themselves.As there can be a significant number of studies, we implemented a filter and search for studies in the user's personal account, as well as sorting options based on the parameters in the study table. This solution allowed for optimizing the process of searching within a large dataset for individual users;
  2. Understanding how image annotation is practically carried out, the tools specialists use during the annotation process, and researching analogous tools helped our team prioritize tasks related to the development of annotation tools.We found that in order for a radiologist to provide a conclusion, images need to be annotated by two specialists. Consequently, a technical task arose for dividing and displaying the annotations of two users within one study;
  3. We investigated the necessity of this development since there wasn't a clear understanding of its significance in the initial stages. We determined that the primary target audience for the generator would be novice specialists, students, and educators. The importance of this development lay in creating complex medical cases while maintaining patient confidentiality and refraining from using real images.An important aspect was studying the possibility of combining multiple pathologies on a single lung image. Through consultations with experts in this field, we discovered that certain pathologies cannot coexist on an image, as some diseases are mutually exclusive of each other.

The developed interface allows doctors to create artifacts that can serve as a basis for automating the diagnostic process of studies.

The intuitively user-friendly interface developed significantly simplifies the user-system interaction process, allowing for the optimization of specialists' work time.

Business Value

  1. Reducing the number of medical errors caused by a lack of experience with rare pathologies. The solution allows for the artificial creation of images with rare pathologies and working with them, thereby increasing the level of expertise among specialists.
  2. Decreasing the time spent on working with studies by reducing the time required to adapt to the platform's interface.
  3. Enhancing the quality and speed of study annotation in the future through the implementation of AI solutions trained on datasets created using our platform.
AR Office application
AR Office application
Machine learning model for predicting startup success
Machine learning model for predicting startup success