New achievements of AI medical imaging by the Chinese Academy of Sciences: Artificial intelligence can provide non-invasive classification for liver cancer patients

According to Leifeng.com news, the Suzhou Institute of Biomedical Engineering of the Chinese Academy of Sciences, together with the research teams of Lishui Central Hospital and the Second Affiliated Hospital of Soochow University, recently carried out a new study.

The results of the study showed that the artificial intelligence system SE-DenseNet used in conjunction with medical imaging combined with enhanced MRI images can complete cancer grading for patients under non-invasive conditions. The research team said that it will apply this technology to the liver cancer ablation planning navigation system it developed to assist in more accurate surgical planning.

Among primary liver cancers, hepatocellular carcinoma (HCC) is an important type of liver cancer, accounting for 70% to 90% of primary liver cancers and is the third leading cause of cancer death worldwide.

The grading of liver cancer has important clinical significance for the patient's clinical diagnosis, treatment selection and prognosis.

Unlike most tumors, liver cancer can be diagnosed through non-invasive imaging tests. Currently, the means of diagnosing liver cancer include imaging tests, biopsy, AFP serum tests, etc. The most commonly used medical imaging tests include CT and MRI. CT and MRI have been recognized as the first choice for non-invasive examinations of diseases such as hepatobiliary and breast cancer.

Pathological biopsy is still a necessary means to assess the malignancy of lesions. If lesion grading based on medical images can be achieved, it can provide reference opinions on tumor treatment plans to a certain extent, reduce the dependence of diagnosis on pathological biopsy, and greatly alleviate the pain of patients.

However, in clinical applications, the grading results are highly dependent on the doctor's experience and are highly subjective. Therefore, seeking an objective and effective grading evaluation method is an important research direction.

With the continuous development of pattern recognition, machine learning, deep learning and other technologies, using medical image-assisted diagnosis systems to build deep learning network models and objectively and automatically grade liver cancer has become one of the current mainstream research directions.

Researchers Dai Yakang, Zhou Zhiyong and Zhou Qing from the Suzhou Institute of Biomedical Engineering of the Chinese Academy of Sciences, together with the team of Vice President Ji Jiansong of Lishui Central Hospital and the team of Director Fan Guohua of the Second Affiliated Hospital of Soochow University, proposed the SE-DenseNet network and carried out a study on the malignancy grading of hepatocellular carcinoma based on enhanced MR images (layer thickness ranging from 3mm to 8mm).

Leifeng.com learned that the study obtained enhanced MRI images of 75 patients from Lishui Central Hospital and the Second Affiliated Hospital of Soochow University, including 75 arterial phase images, 75 venous phase images, and 63 delayed phase images, with a total of 213 lesion ROIs.

The researchers built the SE-DenseNet network by combining the two network structures of DenseNet and SENet in deep learning, and used SENet to self-learn the weights of features to achieve the purpose of enhancing important features. To a certain extent, SE-DenseNet alleviated the feature redundancy of DenseNet.

_{SE-DenseNet framework diagram}

Experimental results show that the classification performance of SE-DenseNet is better than that of DenseNet and DenseNet-BC (SE-DenseNet: accuracy = 0.83, DenseNet: accuracy = 0.72, DenseNet-BC: accuracy = 0.66).

The researchers said that the artificial intelligence system SE-DenseNet used in conjunction with medical imaging combined with enhanced MRI images can complete cancer grading for patients under non-invasive conditions. In the future, this technology will be applied to the liver cancer ablation planning navigation system they developed to assist in more accurate surgical planning.

Zhou Zhiyong, a researcher at the Suzhou Institute of Biomedical Engineering who participated in the study, once said, "Compared with traditional cancer grading by puncture, the use of 'medical imaging + AI' grading can obtain more comprehensive lesion information and reduce the probability of missed detection. In recent years, the accuracy of lesion grading using artificial intelligence has been continuously improved, indicating that this technology has a broad prospect for application in disease diagnosis and treatment."

In addition, Leifeng.com learned that this research was also funded by the National Key R&D Program, Zhejiang Province Key R&D Program, and Suzhou People's Livelihood Science and Technology projects.

Alibaba pays more than 140 million yuan in taxes every day on average; Final judgment on Samsung Note7 explosion case

Android flagships are all equipped with rear-mounted 3D cameras to compete with Apple's first release

Apple's entire product line has reduced prices; 35 universities have added AI undergraduate majors; Yu Chengdong responded to Huawei's decision not to go public

The second batch of acceptance lists for the Science and Technology Innovation Board were announced. Why were all AI unicorns “annihilated”?

Zhou Hongyi launched three new products, targeting middle-aged people who are married and have children and are particularly tired of living.