Long-bone fracture detection using artificial neural networks based on line and contour features of X-ray images

Simple item page
dc.contributor.authorYang, Alice Yi
dc.date.accessioned2019-11-12T09:20:03Z
dc.date.available2019-11-12T09:20:03Z
dc.date.issued2019
dc.descriptionA dissertation submitted to the Faculty of Engineering and the Built Environment, University of the Witwatersrand, Johannesburg, in fulfillment of the requirements for the degree of Master of Science in Engineering. Johannesburg, May 201en_ZA
dc.description.abstractExisting Computer Aided Diagnosis (CAD) systems traditionally employ Convolutional Neural Networks (CNNs) for the classification of particular conditions in the medical field. The objective of CAD systems is to provide a second opinion before a final decision is made by medical physicians. Although existing CAD systems obtain high accuracies, the cost for high accuracy is an extensive amount of images required for the system training. The aim of this research is to achieve high accuracy with a reduced number of required images for training through the use of line and contour-based features. This leads to the proposed question “Can the use of pattern recognition in the form of Artificial Intelligence detect long-bone fractures based on line and contour features from X-ray images in a similar manner as a medical professional?”. The proposed question is answered with the development of two novel schemes for both line and contour-based features. There are two line-based fracture detection schemes, which are Standard line-based fracture detection and Adaptive Differential Parameter Optimization (ADPO) line-based fracture detection. The ADPO scheme optimizes the parameters of the Probabilistic Hough Transform for detailed fracture line detection. The classification of the fractured and non-fractured lines is performed using an Artificial Neural Network (ANN). A total of 13 features are extracted from each detected line. The evaluation of the Standard scheme indicated an average accuracy of 74.25%, whilst the ADPO scheme achieved an average accuracy of 74.4%. The contour-based fracture detection schemes are based on the line-based fracture detection scheme to further improve the classification accuracy. The two contour-based fracture detection schemes are Standard Contour Histogram Feature-Based (CHFB) and improved CHFB fracture detection schemes. A total of 19 features are extracted from each detected contour. The Standard CHFB scheme achieved an average accuracy of 80.7%, whilst the improved CHFB scheme achieved an average accuracy of 82.98%. Both the contour-based fracture detection schemes perform significantly better than the line-based fracture detection schemes. Additionally, a hierarchical clustering technique is implemented to highlight the fractured region within the X-ray image.en_ZA
dc.description.librarianMT 2019en_ZA
dc.identifier.urihttps://hdl.handle.net/10539/28419
dc.language.isoenen_ZA
dc.titleLong-bone fracture detection using artificial neural networks based on line and contour features of X-ray imagesen_ZA
dc.typeThesisen_ZA
Files
Original bundle
Now showing 1 - 1 of 1
No Thumbnail Available
Name:
Alice_Yi_Yang_Masters_Dissertation.pdf
Size:
4 MB
Format:
Adobe Portable Document Format
Description:
Download
License bundle
Now showing 1 - 1 of 1
No Thumbnail Available
Name:
license.txt
Size:
1.71 KB
Format:
Item-specific license agreed upon to submission
Description:
Download
Collections
ETD Collection