Ping An AskBob Doctor’s Deep Learning Model Improves Diagnosis of Kidney Disease

0
50

Ping An Insurance (Group) Company of China, Ltd. (hereafter “Ping An” or the “Group”, HKEx:2318; SSE:601318) is pleased to announce that its smart healthcare team[1] has joined the National Clinical Research Center of Kidney Diseases in China, to present its analytic renal pathology system (ARPS) in a paper published in the Journal of Pathology[2], a respected global peer-reviewed medical journal. The system uses deep learning, a type of machine learning, to identify certain types of kidney lesions and cells to help doctors diagnose kidney disease with greater accuracy.

The ARPS is currently being trialed in house by Ping An through AskBob Doctor, an artificial intelligence (AI)-based diagnosis and treatment assistant tool developed by Ping An. AskBob Doctor, launched in June 2019, has been used 33 million times by 430,000 doctors for smart medical decision support, smart imaging recognition, diagnosis follow-up and patient education.

The ARPS focuses on the glomerulus, a tuft-like group of small blood vessels in the nephron, the structural unit in the kidney that filters blood and produces urine. There are about one million nephrons in a human kidney. Identification of glomerular lesions and their structure is critical for diagnosis, treatment and prognosis evaluation in kidney disease. The process is time consuming and requires a very accurate quantitative analysis method. In the research, 400 Chinese patients with immunoglobulin A nephropathy (IgAN) were studied, with a comparison of the performance of the ARPS and four junior pathologists.

The study found that the ARPS identified three kinds of intrinsic glomerular cells with 92.2% accuracy, which was 5% to 11% more accurate than the pathologists. With sufficient training, the ARPS could obtain an average precision of 0.882 for intrinsic glomerular cell recognition, surpassing the pathologists by 3% to 15%. It took only 0.6 seconds for the ARPS to estimate the number of intrinsic cells per glomerulus, 50 to 90 times faster than the pathologists. The ARPS also achieved an extraordinary performance for glomerular lesion identification and classification with an average accuracy of 92.8%.

Metric

ARPS

Doctor 1

Doctor 2

Doctor 3

Doctor 4

Precision

0.882

0.763

0.764

0.828

0.856

Accuracy

0.922

0.810

0.814

0.874

0.882

Time taken

0.6 s/image

32-54 s/image

Source: Identification of glomerular lesions and intrinsic glomerular cell types in kidney diseases via deep learning, Journal of Pathology, 16 June 2020

Dr. Xie Guotong, Ping An Group Chief Healthcare Scientist, said: “Pathology sets a gold standard for disease diagnosis and treatment. It takes several decades to train a qualified pathologist and it is a great challenge to conduct accurate quantitative analyses due to the high complexity of pathological images. The ARPS can be employed in training and improving the level of pathologists in glomerular lesion identification and effectively improve the diagnostic speed and accuracy of junior pathologists. We hope our work can pave a path for highly accurate pathology-based diagnosis, prognosis, treatment and scientific research.”

Ping An Group, driven by its “finance + technology” and “finance + ecosystem” strategies, has made great strides in healthcare technology development and innovation. Ping An’s smart healthcare team has provided an AI technology platform to more than 17,000 medical institutions across 90 cities in China up to July 2020.

LEAVE A REPLY

Please enter your comment!
Please enter your name here