Applying Deep Learning to Breast Lesion Ultrasound – Another World-First
Samsung Electronics has a very large portfolio of products across various markets, and one thing Samsung does very well is leveraging its vast array of technologies and expertise to each of its products.
For example, Samsung introduced color LCD screens to mobile phones (2002); it was also the first to sell digital televisions (1998, back in the days, TVs only had analog inputs). Samsung also offers printers with an array of mobile and cloud printing functions and aninterface using Android tablets.
This philosophy goes for Samsung’s Health and Medical Equipment Business as well. A few years ago, Samsung won a design award for the user interface (UI) that applied touch screens on x-ray devices, which innovated the way radiologists interact with the machines. This year, Samsung upgraded its RS80A with Prestige ultrasound device with several new features including ‘S-Detect for Breast,’ which applied deep learning to ultrasound diagnosis for the first time in breast lesion analysis.
Recognition Technology, Image Processing and Big Data Analytics
There aren’t many computer-aided diagnosis (CADx) solutions for ultrasound because ultrasound images are read in real-time and it’s very difficult to build a solution that can process the images fast enough. Also, ultrasound images traditionally have relatively more noise and lower resolution compared to other imaging tests, making it harder to accurately process the images.
However Samsung has long developed advanced image processing techniques at theSamsung Advanced Institute of Technology (SAIT). The Institute also researches various future technologies including high-performance computing, big data analytics and recognition technology. So, the decision was made to integrate the fruits from Samsung’s extensive R&D to address this opportunity. The CADx solution for ‘S-Detect for Breast’ is another one of the plethora of SAIT’s technologies that have been applied to Samsung’s products.
Building an Algorithm Based on Deep Learning
A lot of cutting edge hardware is required to build a good deep-learning system. For example, AlphaGo, which made the headlines defeating professional player Lee Sedol in the strategic board game ‘go,’ used racks of custom hardware dedicated to machine learning. While it would be great to have this kind of computing power available for each ultrasound device, the cost wouldn’t be feasible.
Another option could be to feeding patient diagnosis information into a designated server to be processed, but this client-server model poses many potential privacy and security issues. There are also many legal restraints on sharing patient information outside the medical institute.
So instead of building overly expensive ultrasound devices or information systems, Samsung decided to create an algorithm that could be run at each individual device, based on deep learning from an independently acquired anonymous set of real-life data.
With partnering medical institutes and doctors, Samsung put together a database of anonymous breast ultrasound images of patients who agreed to do so. Then Samsung built deep learning system – a very powerful computer that could process the images, and fed the large set of data it acquired so the machine could learn how to better analyze breast lesions.
Based on the 10,000+ images and the decades-worth of experience from partnering hospitals like the Samsung Medical Center, Samsung came up with an algorithm that would learn to detect breast lesions, help determine the size and shape, give advice on whether it is benign or malignant, and integrate all that information into a single report.
While final diagnosis and decisions are made by doctors, S-Detect for Breast now equip them with Samsung’s latest technologies to help them make better decisions faster.
Eyes on the Future
The RS80A with Prestige isn’t just another ultrasound device, but a product of Samsung’s imagination; introducing its cutting edge technologies to different fields. With the RS80A with Prestige, Samsung aims to aid radiologists in making quick and accurate diagnosis, hopefully preventing unnecessary MRI scans and biopsies.
Building on its achievements, Samsung will further work with partnering medical institutes to further improve the algorithm and hopefully, we’ll be able to expand the application of this technology to other areas such as thyroid, liver, etc.
To see how RS80A with Prestige works, see: https://youtu.be/gfw5Qex6Ln8.
More information on RS80A with Prestige can be found at:http://www.samsungmedison.com/micro/RS80A_with_Prestige/index.jsp.
For more on Samsung’s Health and Medical Equipment, visit:http://www.samsung.com/global/business/healthcare.
* Updated to explain Samsung Advanced Institute of Technology’s role in the project.
Posts
No comments: