Lin Ling

Tianjin University, T’ien-ching-shih, Tianjin Shi, China

Are you Lin Ling?

Claim your profile

Publications (4)5.45 Total impact

  • Source
    [Show abstract] [Hide abstract]
    ABSTRACT: Noninvasive determination of tissue optical properties is essential for clinical applications in medical diagnostics and therapeutics. In this paper, we describe a method to determine reduced scattering coefficient μs' and absorption coefficient μa from spatially resolved relative diffuse reflectance. A neural network in conjunction with data preprocessing technique - principal component analysis (PCA) is employed to perform the estimations from the diffuse reflectance data generated by Monte Carlo simulations. The PCA-NN was trained and tested on the space with μs' between 0.1 and 2.0 mm<sup>-1</sup> and μa between 0.01 and 0.1 mm<sup>-1</sup>. Tests on the above space show the rms errors of this method to be 4.6% for μs' and 9.2% for μa. It's a more efficient and robust way for real-time clinical applications than other methods.
    Neural Networks and Signal Processing, 2003. Proceedings of the 2003 International Conference on; 01/2004
  • Ye Wenyu, Li Gang, Lin Ling, Yu Qilian
    [Show abstract] [Hide abstract]
    ABSTRACT: A new method for clustering analysis of QRS complexes is proposed. The method integrates principal component analysis (PCA) with self-organizing map neural network (SOM). The QRS complex feature is extracted based on PCA and the unsupervised SOM is employed to cluster the data. The characteristics and the behavior of the proposed method applying different SOM architectures are studied. The method is tested with the MIT-BIH database. It is demonstrated that QRS complexes feature can be presented by four largest principle components and the PCA results can be used to cluster analysis efficiently. The relationship between cluster results and clinical categories are also investigated in this paper.
    Neural Networks and Signal Processing, 2003. Proceedings of the 2003 International Conference on; 01/2004
  • [Show abstract] [Hide abstract]
    ABSTRACT: The issue we address in this article is how to reduce the computational burden by using an algorithm based on a linear-approximation distance-thresholding compression technique combined with the backpropagation neural network method. We also address how to improve the training speed. The experimental results found with the MIT-BIH database show that the new algorithm is faster in convergence and more accurate in feature recognition than existing methods.
    IEEE Engineering in Medicine and Biology Magazine 04/2000; · 2.73 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: The system clock in the authors' Intel 80c31-based Holter recorder is 12 MHz, and thus it takes tens of milliseconds to obtain the d/sub k/ at instant k. Therefore, it is impossible to apply the original linear approximation distance thresholding (LADT) algorithm in real time. With the authors' improved algorithm it takes about one ms to calculate /spl verbar/a/sub adm//spl verbar/=/spl sigma//spl middot/(1+(a/t)/sup 2/)/sup 1/2/. Though it is much faster to fulfill a linear segment approximation by the improved algorithm than by the original one, it is still difficult directly to apply the improved algorithm in a real-time system. However, it takes only two microseconds to get /spl verbar/a/sub adm//spl verbar/ by look-up in the error threshold table. In this way, about one millisecond is needed to complete a linear segment approximating (about 50 points). It can be seen that the fast realization scheme using the error threshold table not only preserves the high compression ratio and low distortion advantages of the LADT algorithm, but also makes possible the application of the LADT algorithm in real-time.< >
    IEEE Engineering in Medicine and Biology Magazine 05/1994; · 2.73 Impact Factor