 |
| Principal engineers Yeong Kyeong Seong (left) and Moon Ho Park--Courtesy of Samsung |
Samsung has applied deep learning to breast lesion analysis via ultrasound for the first time, the company claims. The algorithm it has developed is already incorporated into an upgraded version of its RS80A with Prestige ultrasound device.
Creating a computer-aided diagnostic solution for ultrasound is particularly challenging because ultrasound images are read in real time, so it's difficult to build a fast enough approach. In addition, ultrasound images have a lot of noise and lower resolution, making them more difficult to accurately process.
Led by principal engineers Yeong Kyeong Seong and Moon Ho Park, the team drew on more than 10,000 anonymous images from partner hospitals including Samsung Medical Center in Seoul, South Korea to create an algorithm that would help to detect breast lesions, determine a lesion's size and shape and even advise on whether it is benign or malignant.
Known as S-Detect, the report offers standardized analysis and classification of suspicious lesions. It includes three modes so that users can set the level of sensitivity and specificity necessary.
 |
| S-Detect for breast--Courtesy of Samsung |
Breast ultrasound is often used to detect further information after a breast change is detected on a mammogram. It can tell the difference between fluid-filled cysts and solid masses, with the former typically not being cancerous. Ultrasound can also be used to guide a biopsy needle. It's more widely available and less expensive than other breast cancer screening options such as MRI.
Rather than create a cloud-based solution, they opted to create an algorithm, which will be updated subsequently, that can run on an individual device. That approach avoids the potential privacy and security issues that a server model would raise.
Samsung is working with its partner medical institutes to further improve the algorithm. It hopes to expand the application of it to other areas including the thyroid and liver.
- here is the announcement