Semester of Graduation
Master of Science in Engineering Science (MSES)
The problem of predicting wrist fractures from X-rays using Artificial Intelligence (AI) methods is addressed. Wrist fractures are the most commonly misdiagnosed fractures because of the complex anatomical structure of the wrist bone which includes several different bones. This research provides a predictive solution to automate the process of wrist fracture classifications and outlines a visualization technique to identify the probable location of the fractured region on the X-rays. This thesis describes a deep learning based approach for wrist fracture classification. Deep convolutional neural network (CNN) based models have been used for wrist fracture classification by combining different optimization techniques. The concept of transfer learning has also been incorporated to investigate if the deep models pre-trained on non-medical images can be used to improve the accuracy for wrist fracture classification task. Advancements in the automation of wrist fracture detection rate using transfer learning approach is reported. After training an ensemble of models, the accuracy of the proposed technique has been validated with the help of senior orthopedic surgeons and radiologists using the developed visualizations that provide probable locations of the fractured region on X-rays. The reported results provide promising solutions that can help reduce the misdiagnosis of wrist fractures, enhance patient care, and help facilitate improvements in patient morbidity.
Thomas, Dineep, "Artificial Intelligence Based Wrist Fracture Classification" (2019). LSU Master's Theses. 4991.