Factsheet

ScanDiags will be the world’s first product to interpret MRI of all body regions for orthopaedic conditions, providing AI-driven augmented diagnosis to clinicians and radiologists. The project was initiated in 2016, after preliminary work since 2011. In May 2019, the fifth generation prototype was made available.

A coordinated effort of international health providers

The ScanDiags service architecture

In cooperation with a diverse group of hospitals, a comprehensive catalog containing the majority of MRI-identifiable clinical conditions is being AI-trained. The ongoing
engineering process will continuously add new modules from this catalog to the ScanDiags product.

The ScanDiags service can run locally in a customer’s internal IT infrastructure or in the cloud. It will receive one or more MRI-sequences via a direct, protected and monitored connection from either an MRT machine, PACS system, RIS, other software layer or via direct user interaction through the ScanDiags test app or website. It communicates with PACS, RIS, EHR and other systems through standardized APIs, such as DICOM and HL7. The service will optionally call into EHR systems to gather additional data, or look up its own internal historical data for the analysis of time sequences. The service processes all input data and performs its interpretation and prediction routines. After concluding, the IT returns its recommendations to the calling system or user as augmented diagnosis or prediction.

ScanDiags’ artificial intelligence

ScanDiags is a modern implementation of artificial intelligence.
It combines deep learning, image and text analysis, as well as regression-based machine learning methods. Its core AI is built by applying supervised and unsupervised deep learning methods on unstructured MRI-images and structured EHR-data. Data is cleansed, pseudonymized and preprocessed through traditional image analysis and data aggregation methods. Image data used for training the ScanDiags AI is labeled through automatic text extraction from clinical reports and augmented with manual validation by experienced
radiologists. No annotation of image content is applied, and a broad and heterogeneous set of data sources from different clinical practices, MRT manufacturerplatforms and sequence parameter values is aggregated into the AI training process. The ScanDiags AI thus learns autonomously and with as minimal bias as possible. All used data sets are backtrackable for quality validation and GDPR-compliance.

Continuous learning

Finalized reports from radiologists and clinicians will be reprocessed by ScanDiags periodically. This feedback loop allows for ScanDiags’ continuous learning and quality-improvement.

Globally aggregated radiology-expertise

ScanDiags, through its feedback loop and redistribution of virtualized radiology skills, will provide consolidated knowledge and experience from all globally participating institutes, radiologists and clinicians.

AI-driven augmented diagnosis and prediction on musculoskeletal conditions from automated interpretation of MRI

ScanDiags automatically interprets sequences of musculoskeletal MRI images.
Historical and current EHR-information can be aggregated for monitoring and prediction.
Use cases include real-time second opinion, peer review, retrospective review, case comparison and more.
Catalog of detectable conditions is being developed in cooperation with Switzerland’s leading hospitals.

Clinical trials since 2018 show AUC, specificity and sensitivity on par with experienced human radiologists.
ScanDiags’ AI-algorithms perform equally well independent of MRT manufacturer and image parameters.
Large and heterogeneous sets of training data from many different sources avoid bias and provide quality.

Improve detection rate for rare conditions through AI-consolidated radiology knowledge.

Avoid repetitive tasks, e.g. when qualifying and segmenting MRI of the spine.

Prioritize and triage cases automatically during times of high pressure.

Easily look up similar cases from colleagues, even across organizations.

Avoid unnecessary treatments through predictive AI-capabilities.

Have radiologists‘ skills available in remote locations and during off-hours.

Save time by automating peer reviews, knowledge sharing and quality assessments.

Simplify training of new radiologists through automated assessments.

Justify treatment recommendations with automated predictive second opinions.