The results of the Indian Diabetic Retinopathy Image Dataset (IDRiD) fundus analysis competition organized by the IEEE International Symposium on Biomedical Imaging (ISBI) were recently announced. Ping An Technology's PAMIA (Ping An Medical Imaging Assistant) performed head and shoulders above its peers, including a world first in EX (hard exudates) image segmentation, world second in HE (bleeding) segmentation and world third in MA (microaneurysm) segmentation task.
Ping An Technology again wins world-class award in medical field after setting new world record in LUNA rankings
This is the second time this year that Ping An Technology has garnered awards in a world-class competition in the medical imaging research sector, heralding another milestone for the company in artificial intelligence. Earlier in the year, PAMIA had set a new record for nodule detection with an accuracy rate of 95.1 percent, and beat its own previous record for false positive reduction with an accuracy rate of 96.8 per cent in the international authoritative Lung Nodule Analysis (LUNA) rankings for medical imaging.
ISBI is an international conference dedicated to biomedical imaging sponsored by the IEEE Signal Processing Society (SPS) and the IEEE Engineering in Medicine & Biology Society (EMBS). The competition attracted the participation of 22 well-known research and medical institutions, among them, iFlytek, Nanyang Technological University, National Tsing Hua University, Peking University and Samsung. The diabetic retinopathy (DRP) segmentation and classification challenge using the IDRiD dataset included three subsections: lesion segmentation, case classification and monitoring of the optic disc and central fovea. The subsection that the PAMIA platform participated in is the most difficult competition in terms of lesion segmentation and best reflects the strength of intelligent film reading.
DRP is one of the most common microvascular complications in diabetes, with a very high incidence of vision impairment including, in some cases, total blindness. Timely detection and treatment can greatly reduce the risk of vision loss, while early detection with an indication for treatment has important clinical significance. The three lesions, EX, MA and HE, are key to the diagnosis of DRP, in which EX is the main characteristic of the middle stage of the disease, while also signaling the critical time period for engaging in a treatment regimen. The detection and segmentation model for EX, HE and MA can be used to display the edge of the relevant lesions on the basis of classification prediction of DRP, assisting doctors in diagnosing and deciding on the best treatment options quickly. PAMIA ranks first worldwide when it comes to EX, the most difficult yet most important lesions segmentation challenge, while ranking among the top three globally for the segmentation of other lesions.
The fundus image is different from other image scenarios, as these lesions only occupy a few dozen pixels. PAMIA, in order to ensure the detection sensitivity, combined its comprehensive advantages in the field of medical diagnosis and deep learning with its experience of setting new world records in the LUNA competition at the beginning of the year to design a new segmentation network that used deep and transfer learning theories to identify suspected lesions quickly and achieve boundary segmentation. At present, the marking and segmentation of lesions can be completed in a few seconds, greatly improving the reading speed and accuracy rate.
Ping An Technology has achieved several milestones in the intelligent film reading field since 2016 when it entered the medical imaging sector. These achievements reflect the company's leadership and dominant position in the technologies related to medical imaging. With this award as a new milestone, PAMIA plans to devote itself to furthering its research on medical artificial intelligence, helping to achieve a strategic development goal for China's artificial intelligence sector.
0 comments:
Post a Comment