Recently, it was learned that the technology research institute, the “Zhang Guofan Lung Microjunction Diagnosis and Treatment Center†affiliated to Fudan University’s Huadong Hospital and the “SJTU-UCLA Machine Sensing and Reasoning Joint Research Center†of Shanghai Jiaotong University formed a joint research project “3D Deep Learningâ€. From CT Scans Predicts Tumor Invasiveness of Subcentimeter Pulmonary Adenocarcinomas was published in the American Association for Cancer Research (AACR) journal Cancer Research, which had an impact factor of 9.13 in 2017.
Paper screenshot
The article was published online October 2, 2018. The article uses a deep learning approach to train pixel-level subcube lung adenocarcinoma CT data and its case outcome annotations through a multitasking convolutional neural network to sub-cm lungs. The degree of invasive risk of adenocarcinoma was automatically predicted before surgery, and the learning difficulty, migration generalization ability, stability and reliability of the task spectrum reduction model on medical imaging were established. This research can help doctors choose the treatment of early lung cancer, which will effectively promote the development of precision medicine .
CT imaging predicts early tumor infiltration and accurately solves the problem of lung cancer screening
Professor Li Ming from Huadong Hospital of Fudan University talked about the status quo of lung cancer: “China is a country with high incidence of lung cancer. The 5-year survival rate is less than 20%, and the mortality rate ranks first among all cancers. The reason is due to the lack of domestic patients. Early screening consciousness, patients often find lung cancer in the middle and late stages, and the treatment is weak at the current medical level. At the same time, the high medical expenses not only make the patients unable to make ends meet, but the corresponding medical insurance also imposes a huge burden on the country. China has issued a number of policies to try to decentralize patients to the grassroots level, a process that requires artificial intelligence to assist."
However, there are many domestic lung nodule companies. Although the accuracy of image recognition is almost the same, the whole diagnosis process is mixed. In order to stand out in the whole industry, the intra-site technology tries to divide the lung nodules into four sub-categories of AAH, AIS, MIA and IA in a multi-classification manner, giving suggestions on the degree of early infiltration, and further exploring the pulmonary nodules of patients. Case.
On the 128 test sets, the multi-task deep learning model predicted better results than the four radiologists (two senior physicians and two low-grade doctors). The accuracy of the model in distinguishing between infiltrating/non-invasive classifications reached 78.8% (AUC), and the accuracy of distinguishing IAC/non-IAC (phase 0/I) classifications reached 88.0% (AUC). The accuracy of distinguishing the AAH-AIS/MIA/IAC classifications reached 63.3% (F1).
Most of the sub-cm lung nodule data used in this study were lung-milled glass nodules. This type of nodules, especially sub-cm-milled glass nodules, showed less invasive images on CT images due to traditional malignant signs. The pre- and post-invasive lesions have high overlapping images, and the diagnosis is very difficult. In the diagnosis of the three classifications, the diagnostic accuracy rate of the senior physician is only 56.6%, and the depth learning accuracy within the point can reach 63. .3%, which shows the advantages and prospects of deep learning in dealing with such problems.
From conception to publication, this article has experienced data acquisition, pixel-level annotation, data processing, model development training, model testing, public data set application, download, labeling, testing, paper writing, modification, peer review, repair and other processes. The joint research team within the point only completed the algorithm development test and paper publication in less than 9 months.
China Extract Powder For Use As Dietary Supplement Extract Powder, Extract Powder Manufacturer
Shaanxi Kang New Pharmaceutical co., Ltd. , https://www.apipepdites.com