Local Time : 06:05 CET

C-2539 - Predicting smoking status using deep learning methods on brain MRI images

S. Wang, R. Zhang, Y. Deng, K. Chen, T. Jiang, D. Xiao, P. Peng; Beijing/CN Type: Scientific Exhibit
Area of Interests: Neuroradiology brain, CNS
Imaging Techniques: CAD, MR
Procedures: Structured reporting, Decision analysis, Computer Applications-Detection, diagnosis
Special Focuses: Image verification, Outcomes
Add poster to my schedule In your schedule (remove)

Aims and objectives:  Smoking is a major public health problem, and the world‘s leading cause of preventable deaths. Study confirms that nicotine dependence lead to brain structure changes, but the change is subtle, the clinical doctors cannot find and interpret the change of brain MR images and the degree jud[...]

Methods and materials:  The head MRI images of 61 smokers and 66 non-smokers were enrolled, there are 176 images in each subject 25 percent of smokers and controls were randomly selected as the test set (15 smokers and 16 non-smokers) the rest subjects were used as the training set.   Two deep learning model wer[...]

Results: The Conv3D depth model of smoking classification accuracy reached 80.6%, the sensitivity was 80.0%, the specificity was 81.3% after many experiments in our research. The ConvLSTM depth model accuracy of smoking classification reached 93.5%, the sensitivity was 93.3%, the specificity was 93.75%.[...]

Conclusion:   The discriminant accuracy of the two machine learning models which used in our research was significantly higher than the reported result of using the SVM classification. The deep learning model maybe embody complex biomarker that could be used to evaluate treatment efficacy.   [...]

Personal information:

References: 1. Gu D, Kelly TN, Wu X, et al. Mortality attributable to smoking in China. N Engl J Med 2009; 360: 150-159 2. Yang GH, Ma JM, Liu N, et al. [Smoking and passive smoking in Chinese, 2002]. Zhonghua Liu Xing Bing Xue Za Zhi 2005; 26: 77-83 3. D. Tran LB, R. Fergus, L. Torresani, and M. Paluri. Learni[...]

This website uses cookies. Learn more