fracture.ai uses smart computer vision techniques to detect bone fractures in your X-ray examinations. The image feature extraction and machine learning algorithms were trained on a large database. By using fracture.ai you help to further increase our fracture detection rate!
When you upload an X-ray study to the server, the images will be analyzed by our algorithms, based on neural networks computing image classification and object detection tasks.
Radiographs of injured body regions are commonly performed to diagnose fractures. They still are the first line of imaging diagnostics in many acute trauma indications. Our algorithms are trained to detect fractures in X-ray examinations. By using fracture.ai, you provide us with additional X-rays to improve the sensitivity of the technique!
You need a modern web browser (like Microsoft Edge, Google Chrome, Mozilla Firefox, or Apple Safari) to be able to properly use the website. Please make sure to use an up-to-date version of your web browser.
Additionally, You need to be in the legal position to upload content You want to be analyzed according to the Terms of Service, Privacy Policy, and Disclaimer.
Neural networks need to be trained on large amounts of data to achieve accurate and
robust results. You can upload your DICOM or IMAGE files to our servers to enable
detection of bone fractures. When doing so, You agree to grant fracture.ai a license on
Your anonymized X-ray file(s) under the Creative Commons
CC0
license.
With the help of your cases, we improve our service for You and all other users!
Data protection is a very important topic for us.
As long as you upload DICOM or other format X-rays not containing visible text, you are providing anonymized data. This means that it is impossible to re-identify an individual person.
We regularly audit received images regarding a potential presence of personal data. Should there be any doubt about an image containing sensitive information, we will delete it as soon as detected
Do not upload files with personal information written on the image!
We strip your DICOM file's metadata inside your web browser. Only the image pixels are uploaded to the servers. DICOM header information is used to create a suitable filename, which consists of the hashed patient ID, a hashed timestamp, gender, and rounded age. For detailed information please refer to our Privacy Policy.
There is a common computer related principle: "Garbage in, garbage out". Results will be best with high resolution input images, featuring sufficient sharpness and contrast. DICOMs are suited best, because they feature full resolultion and more grayscale levels than visible on screen.
vs.
Please also make sure not to upload multiple images on a single canvas. An example is displayed below.
We are working on automatically splitting a canvas containing multiple projections. However, the feature is not finished yet.
fracture.ai typically provides a probibility for a given image to contain a fracture, together with a color-coded heatmap. You should be aware that this heatmap could, but not necessarily does highlight the site of fracture(s).
We perform classification tasks with EfficientNet models, reaching high levels of accuracy varying with body region and age group. An overall algorithm serves as fallback in cases where suitable region- or age-specific models are unavailable.
Model accuracy for different regions and age groups is given in the figure below.