LivGuard identifies signs of cirrhosis in liver ultrasound images. The system is designed to be user-friendly for health professionals irrespective of their level of training. It operates by processing uploaded images with a deep neural network, pinpointing regions likely to be cirrhotic. The technology has the potential to expedite cirrhosis detection. Moreover, it could increase the availability of medical imaging expertise, particularly in areas with limited access to specialized hepatologists.
What’s new: Engineered on deep neural networks, Topazium implemented a framework which detect signs indicative of cirrhosis, one of the most common diseases in healthcare, on a liver ultrasound.
Key insights: Our system accurately identifies and localizes key ultrasonographical findings of cirrhosis in liver ultrasounds even if used by non-trained health professionals.
How it works: Liver ultrasounds images from patients suspected of cirrhosis are uploaded into Topazium’s platform. This input is encoded and processed by a calibrated convolutional neural network that will identify the cirrhotic-likely regions. Based on this, the system returns the probability of a particular image as being classified as positive or negative for cirrhosis.
Results: The system can detect images with evidential ultrasound signs of disease, whilst correctly categorizing those deemed as “negative”. Accuracies are comparable to expert hepatologists.
Why it matters: A liver ultrasound is a safe, non-invasive procedure that produces images of a person’s liver. A liver ultrasound can help detect various conditions that affect the liver, such as cirrhosis. Automated liver ultrasound interpretation could provide substantial benefit in many medical settings. At the level of practicing hepatologists, it can reduce time to cirrhosis detection. It can also increase access to medical imaging expertise in areas of the world where access to skilled hepatologists is limited.
HOW TO RUN THE EXPERIMENT
Access: available through a dedicated app and/or via an API.
Image upload: Upload your image by clicking the “Browse” button. The system can work with any type of image format (jpeg, tiff, png).
Running your experiment: Expected run time for each image: less than 45 seconds
Output: A similarity score that represents the probability (in %) of the image to be classified as normal or abnormal (cirrhosis).