Shulou(Shulou.com)06/03 Report--

Since the outbreak, a number of technology companies have joined the battle against the epidemic.

Among them, screening epidemic situation is the most important part of this campaign, and CT image has become an important basis for screening and diagnosis and treatment of new crown pneumonia.

However, in the critical period of current epidemic diagnosis and treatment, the total number of stock patients and new patients is huge, doctors need to follow-up and compare multiple CT images of patients at different stages, so as to accurately evaluate the patient's disease development and treatment effect.

With traditional visual inspection of medical imaging methods, doctors not only have a huge workload, but also have difficulty in accurately assessing and comparing patients 'conditions in time.

Under the condition that the medical resources of the whole society are tight to fight the epidemic situation and doctors are overloaded, excessive CT imaging examination poses a huge challenge to the front-line anti-epidemic work.

In fact, AI is used for medical image analysis and auxiliary diagnosis, and there are already many application cases. Just yesterday (2.28), Baidu teamed up with Lianxin Medical, officially launched "CT image-based pneumonia screening and disease pre-assessment AI system," and has been put into use in Hunan Chenzhou Xiangnan University Affiliated Hospital.

From minutes to seconds.

It is reported that the CT images of a patient with new crown pneumonia are about 300, which puts great pressure on doctors for clinical diagnosis. It takes about 5-15 minutes for doctors to analyze the CT images of a case.

The AI pneumonia screening and pre-evaluation system, which first landed in the Affiliated Hospital of Xiangnan University, can complete the calculation and display of a complete set of quantitative indicators such as lesion detection, lesion contour delineation, density distribution histogram of both lungs, number, volume and proportion of lung lesions on CT images of patients within dozens of seconds.

Among them, the detection accuracy and recall rate of the system on the test data set reached 92% and 97%, respectively, so as to prevent missed detection on the basis of ensuring high detection accuracy.

In addition to rapid detection and identification of pneumonia lesions, it provides quantitative evaluation information such as the number, volume and lung proportion of lesions for disease diagnosis. At the same time, the system is supplemented with visual means such as histogram of bilateral lung density distribution and lesion delineation superposition display, which provides qualitative and quantitative basis for clinicians to screen and pre-diagnose pneumonia patients, and improves the diagnosis and evaluation efficiency of doctors.

AI system for pneumonia screening and disease pre-evaluation based on CT images

In addition, the deep learning algorithm model adopted by the system fully trains the collected high-resolution and low-resolution CT image data, which can adapt to the examination data collected by different levels of CT imaging equipment, and is expected to provide effective pneumonia auxiliary pre-diagnosis tools for primary hospitals with limited medical resources and limited medical level.

Second, open source against closure

Many companies also want to build their own pneumonia CT impact analysis model, but the cost of training from scratch is high and cannot be used in time.

For this reason, Baidu and Lianxin Medical have adopted an open attitude and opened the AI model of pneumonia CT image analysis in the above system for the first time in the industry_Pneumania-CT-LKM-PP.

In addition, its pre-training model has also been opened on Baidu EasyDL. Developers can select "pneumonia CT image recognition algorithm" in EasyDL image segmentation model, and a small amount of data training can obtain a model further optimized based on actual scenes.

For developers who want to get started with Pneumonia CT Image Analysis Model (Pneumania-CT-LKM-PP), Baidu also gives detailed tutorials.

1. Define the data to be predicted

If you don't have your own data, you can also practice with demo.dcm medical images provided by Baidu.

Display medical images

2. Load the pre-trained model

Baidu's PaddleHub provides a Module for lesion analysis and lung segmentation, i.e. Pneumonia_CT_LKM_PP, which includes two modules: lesion segmentation and lung segmentation, both of which are based on UNet for a series of optimizations.

3. Forecast

PaddleHub For modules that support one-click prediction, you can call the corresponding prediction API of the module to complete the prediction function.

4. Post-treatment

After some post-processing, lung segmentation results are mapped to the original image, and then the focus segmentation and lung segmentation are merged into one image for visualization.

Results after fusing lung segmentation and lesion segmentation

Code Portal:

https://aistudio.baidu.com/aistudio/projectdetail/289819

III. AI war epidemic, fighting for love

With the accumulation of clinical diagnosis data, the imaging big data characteristics of new crown pneumonia are gradually clear, and it is believed that AI will play a greater and greater role in pneumonia screening field.

According to Baidu, the system will be deployed in Hubei, Chengdu and other hospitals for six consecutive years, and its online version will also be open to hospitals nationwide free of charge, which is conducive to medical personnel to carry out remote consultation and cooperation based on the system, improve the disease diagnosis and treatment ability of grass-roots hospitals, and then is expected to reduce the risk of cross-infection caused by patients in the process of referral and inspection.

It is also expected that more hospitals and algorithm researchers will participate in the research and development of AI-based medical image big data anti-epidemic products, contributing to anti-epidemic clinical research and clinical product research and development.

All together, win the epidemic prevention war.

https://www.toutiao.com/i6798806324206371339/

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.