Publication     Conference     Thesis


   
  Doctors in hospitals manually inspect a large number of x-ray images for fractures.
Manual inspection is tedious and time consuming.
A tired radiologist has been found to miss a fractured image among many healthy ones.
Computer vision system can help to screen x-ray images for suspicious cases and alarm the doctors.
 
  Fractures at femur (thight), radius (wrist), and chest.
 
 

Developing a prototype system for field test in local hospital.
International patent application published under Patent Cooperation Treaty (PCT):
    Method and System for Detection of Bone Fractures, International Publication Number WO
    2005/083635 A1, 9 Sep 2005.
US patent filed.
    Method and System for Detection of Bone Fractures. US 2007/0274584, filed 29 Nov 2007.
 

 


1. Compute Neck-Shaft Angle

            accuracy: 93.5%, fracture detection rate: 61.5%

     Femurs correctly identified as healthy:

     

     Femurs correctly identified as fractured:

     
 

  2. Analyze Trabecular Texture Pattern based on Gabor Filtering

         accuracy: 93.5%, fracture detection rate: 46.2%

      Gabor texture orientation maps of correctly identified healthy femurs:

     

      Gabor texture orientation maps of correctly identified fractured femurs:

     

      Combined performance of methods 1 and 2:
            accuracy: 95.4%, fracture detection rate: 76.9%
 
 

3. Analyze Intesity Gradient Map

        color circle indicating gradient direction

       Intensity gradient map of correctly identified healthy femurs:

            

       Intensity gradient map of correctly identified fractured femur:

       

       Combined performance of all methods:
              accuracy: 98.2%, fracture detection rate: 92.2%
 
 

4. Combination of Various Features and Classifiers:

      Combine neck-shaft angle measurement, Gabor texture orientation map, intensity gradient map,
      and Markon Random Field with various classifiers. So far, Gini-SVM seems to perform better than
      other classifiers that we have tested. The combined method can detect very subtle fractures:

          

       Performance:
              Max rule: accuracy: 98.1%, fracture detection rate: 91.7%
              1-of-3 rule: accuracy: 97.2%, fracture detection rate: 100%

      Can detect radius features:

         

       Performance:
              Max rule: accuracy: 95.9%, fracture detection rate: 95.7%
              1-of-3 rule: accuracy: 85.1%, fracture detection rate: 100%
 

  A/Prof. Leow Wee Kheng, Dept. of Computer Science, National University of Singapore.
Dr. Howe Tet Sen, Dept. of Orthopaedics, Singapore General Hospital.
Dr. Png Meng Ai, Dept. of Diagnostic Radiology, Singapore General Hospital.
Ms. Ee Xian He (M.Sc., RA), Dept. of Computer Science, National University of Singapore.

Ex-students:
Mr. Chen Ying (M.Sc., RA), Dept. of Computer Science, National University of Singapore.
Ms. Vineta Lum Lai Fun (Honours), Dept. of Computer Science, National University of Singapore.
Mr. Dennis Lim Sher Ee (Honours), Dept. of Computer Science, National University of Singapore.
Ms. Ee Xian He (Honours), Dept. of Computer Science, National University of Singapore.
Mr. Tian Tai Peng (M.Sc.), Dept. of Computer Science, National University of Singapore.
Mr. Dennis Yap Wen-Hsiang (Honours), Dept. of Computer Science, National University of Singapore.
Mr. William Sze Wing Kay (Honours), Dept. of Elec. & Comp. Eng., National University of Singapore.
Ms. Toh Beng Keow (Honours), Dept. of Elec. & Comp. Eng., National University of Singapore.
 
  F. Ding, W. K. Leow, and T. S. Howe. Automatic Segmentation of Femur Bones in Anterior-Posterior
    Pelvis X-ray Images. In Proc. Int. Conf. on Computer Analysis of Images and Patterns, 2007, pp. 205-212.

J. Congfu He, W. K. Leow, and T. S. Howe. Hierarchical Classifiers for Detection of Fractures in X-Ray
    Images. In Proc. Int. Conf. on Computer Analysis of Images and Patterns, 2007, pp. 962-969.

 Y. Chen, X. Ee, W. K. Leow, T. S. Howe. Automatic Extraction of Femur Contours from Hip X-ray
    Images. In Proc. First International Workshop on Computer Vision for Biomedical Image Applications
    (CVBIA 2005) (in conjunction with Int. Conf. on Computer Vision, 2005). Y. Liu, T. Jiang, C. Zhang
    (Eds.), LNCS 3765, Springer, 2005, pp. 200-209. .

Y. Chen, Model-Based Approach for Extracting Femur Contours in X-ray Images, M.Sc. Thesis, Dept. of
    Computer Science, School of Computing, National University of Singapore, 2005.

V. L. F. Lum, W. K. Leow, Y. Chen, T. S. Howe, M. A. Png. Combining Classifiers for Bone Fracture
    Detection in X-Ray Images. In Proc. Int. Conf. on Image Processing, 2005.

 S. E. Lim, Y. Xing, Y. Chen, W. K. Leow, T. S. Howe, and M. A. Png. Detection of Femur and Radius
    Fractures in X-Ray Images. In Proc. 2nd Int. Conf. on Advances in Medical Signal and Information
    Processing, 2004, p. 249-256.

 D. W.-H. Yap, Y. Chen, W. K. Leow, T. S. Howe, and M. A. Png. Detecting Femur Fractures by
    Texture Analysis of Trabeculae. In Proc. Int. Conf. on Pattern Recognition, 2004, vol. 3, p. 730-733.

T. P. Tian, Y. Chen, W. K. Leow, W. Hsu, T. S. Howe, M. A. Png. Computing neck-shaft angle of femur
    for x-ray fracture detection. In Proc. Int. Conf. on Computer Analysis of Images and Patterns, LNCS
    2756, 2003, p. 82-89.

T. P. Tian, Detection of Femur Fractures in X-Ray Images, M.Sc. Thesis, Dept. of Computer Science,
    School of Computing, National University of Singapore, 2002.
 

  This project supported by NMRC/0482/2000 and NMRC/0759/2003.

3 July 2016