Set of data analytic tools for breast conserving cancer surgery

Customer

Innovative Swiss digital microscopy startup, a spin-off from the Swiss Federal Institute of Technology (EPFL), that enables a global mapping of a lumpectomy for full margin control with histology-grade confidence in few minutes before closing the incision.

Objective

The main objective of the Client’s product is to let clinicians be one touch-on-the-screen away from visualizing cancerous cell on a surgical specimen immediately during surgery. The set of tools includes production grade GUI application for the ultra-high resolution scanner to visualize tissue specimen, a standalone offline viewer with advanced annotations to work with scanner-produced images and a performance assessment web app to evaluate the level of recognition of breast cancer in surgical specimen shown in the images. Additionally a separate module provides support with converting tiff images into the DICOM format with zero error in validation.

Solution

In the heart of the software tools lies the modern ergonomic GUI application whose goal is to operate scanner hardware and collections of images on a wide touch screen and provide smooth real time performance with gigabyte size images. The app includes smooth real-time pan and zoom commands, image repainting and a snapshot feature. Also the operator has a set of tools to put free line drawings, stick a text note and perform measurements, and additionally multiple snapshots could be made to document important moments of image analysis.

The next sub-project has targeted the development of a stand-alone viewer (for desktop or laptop computers) that works with the images imported from the scanner devices. A key functionality was an editable free-form annotation on the images. Pathologists now can work with this annotations tool in order to delineate precisely cancerous zones on the images. The software allows importing annotations produced by several different pathologists on their off-line viewers in order to define consensus between their corresponding annotations. The resulting consensual annotation will be used to define cancerous/healthy segments of the images that in their turn will be provided as inputs to train the AI-based solution.

Almost simultaneously with the Offline Viewer with advanced annotations tool the Customer has asked the team to work on a separate module to help integrate the scanner into hospitals' IT infrastructure and convert tiff images into DICOM.

As the basis of the protocol, the team has used dcm4che library that provides basic access to attributes, sequences, etc. This is the link between high-level and low-level programming. Now the scanner generates tiff images, and with the help of the module the user can convert them into high quality DICOM files with good tags and zero errors in validation.

Physicians Performance Assessment Tool was developed as a separate complementary product in a form of a quiz operated in a browser with the access by specific link that is used by study physicians to assess their level of recognition of breast cancer in surgical specimen. The tool allows displaying images, providing multiple choices for selecting, storing results and displaying the correct answer.

Results

For the moment the product operates stable and successfully and has been advancing currently with the AI and Machine Learning. On a higher level the company now focuses more on strategic marketing and sales to incorporate the scanner and its software into medical institutions and organizations on a larger scale.

Technologies and tools

Java, Java SE version 8, JavaFX, Spring Boot, MySQL, TypeORM, ImageMagick, H2 Database, Liquibase, Apache FOP, Hibernate, dcm4che, IntelliJ IDEA IDE, Eclipse, Gradle, Dozer, Git, Joi, Angular 8, NestJS (Express, Node.JS), Angular Material

Duration: 2017 - now

Team: 3-4 senior Java/JavaFX developers, 1 QA specialist, 1 senior web developer on a need basis

Cooperation model: TM