Analysis in Abnormal Alignment Vertebrae by Neural Network Algorithm
This research is an application of the neural network (NN algorithm) to medical work. The NN algorithm was used to classify and analyze the alignment of human vertebrae from computerized tomography(CT). The data used in this research, CT images and patient information, was obtained by the support from Suranaree university of technology hospital, Nakhon Ratchasima, Thailand. This study focused on the alignment of vertebrae, especially on cervical vertebrae (C1-C7). The methodology consisted of 2 main steps. The first step was the image processing used to extract the essential characteristic of an image. The 3D CT images were processed by RadiAnt Program and 4 positions were extracted to 2D images, which were anterior, posterior, right, and left side. Ridge detection with various parameters was applied to the 2D images. In the final step, we trained the images processed with Ridge detection by the RapidMiner program using NN algorithm to make the prediction model. The model obtained with the highest evaluation had parameters, training cycles=450, learning rate=0.03 and, momentum=0.9. The accuracy, precision, recall and AUC of the model were 62.22% +/- 18.35%, 62.22% +/- 18.35%, 100.00% +/- 0.00% and 0.528 +/- 0.197, respectively. Keywords - CT Scan Image, Cervical Vertebrae Alignment, Ridge Detection, Neural Network.