Progress in colorectal cancer research based on artificial intelligence
Release date: 2018-06-08
In recent years, artificial intelligence (AI) has developed rapidly. From AlphaGo's continuous defeat of human chess players to the unmanned technology that major auto manufacturers are vying to launch, artificial intelligence has made breakthroughs in many fields. In the major industries of artificial intelligence, the direction of medical plus artificial intelligence has become the focus of attention and has attracted much attention. The application of artificial intelligence in medical treatment can effectively alleviate the current situation of serious shortage of medical resources and imbalance in distribution, improve the operational efficiency of medical systems, and even promote the transformation of the entire medical industry.
Colorectal cancer is the third largest cancer in the world with an annualized incidence rate of 0.41‰ and a mortality rate of 0.15‰. If diagnosed early and treated correctly, the five-year survival rate for patients with colorectal cancer is as high as 90%. Once the cancer cells spread to the colorectal, the patient's five-year survival rate drops rapidly. Therefore, early diagnosis and early treatment of colorectal cancer are extremely important.
Segmentation of colorectal cancer lesions is the basis for preoperative prediction, staging, and efficacy evaluation. Due to the blurred boundary between cancer lesions and normal tissues, it is difficult to achieve accurate automatic segmentation by traditional methods, and there are few relevant international studies. Manual or semi-automatic segmentation methods commonly used clinically are extremely cumbersome, time consuming, and highly dependent on the operator.
Gao Xin, Jian Junming, Xia Wei, et al., of the Institute of Biomedical Engineering and Technology, Suzhou Institute of Chinese Academy of Sciences, proposed a T2 weighted MRI image segmentation method for colorectal cancer based on full convolutional neural network (FCN). The method uses VGG-16 network for feature extraction. A side output module is extracted from the last convolution layer of each module in VGG-16. These edge output modules can dig deep into multi-scale features and generate corresponding outputs. Finally, all of the edge output results are fused to produce the final segmentation result. The experimental results show that the model has high sensitivity (87.85%) and specificity (96.75%) for colorectal cancer tumor segmentation.
The significance of this research is to find a method for segmentation of automatic colorectal cancer lesions by artificial intelligence, which can effectively shorten the time required for the preliminary work of colorectal cancer disease analysis and diagnosis, and greatly reduce the labor intensity of doctors. In addition, this method can be extended to other tumor lesions to accelerate the research progress of related diseases. Related results are published in Australasian Physical & Engineering Sciences in Medicine.
Source: Suzhou Biomedical Engineering Technology Research Institute
Zhuhai Mingke Electronics Technology Co., Ltd , https://www.zhmkdz-electronics.com