ANALYZER SOFTWARE DEVELOPMENT FOR BONE HEALTH STATUS USING SUPPORT VECTOR MACHINE

• Main Author: ARKANTY SEPTYVERGY (NIM : 13311055)- NIKITA PRADNYA PARAMITA SETYAGAR (NIM : 13311072)
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/25803
• Institution: Institut Teknologi Bandung

<br /> <p align="justify"> Osteoporosis is a degenerative disease which is indicated by the occurrence of low bone mineral density and deterioration in skeletal microarchitecture. It was shown in a research conducted by the Research and Development Center of Nutrition, Department of Health, Republic of Indonesia that 2 out of 5 people in Indonesia have a significant risk of developing osteoporosis. Early detection of osteoporosis is highly necessary to avoid the occurrence of bone fracture, which could cause in paralyzation and giving an impact to one’s quality of life. <br /> <br /> In this thesis, an analyzer for early prediction of mandibular condyles density as the region of interest (ROI) has been developed. The image processing, consisted of determining the ROI and noise compensating using background-subtraction method, was used to minimize the offset and optimize the segmentation process, both in a high and low density area. Feature extraction in both condyles was executed based on the fraction from the number of pixels with a density above predetermined threshold in the area to the total area of the ROI. Bone health status classification was performed using support vector machine (SVM) algorithm with k-fold cross validation (KCV). <br /> <br /> Classification and data validation was successfully conducted using SVM algorithm with KCV. Validation was executed using the result of DXA test, i.e. bone health status, from 105 patients of postmenopausal women with 50 – 85 years of age from Bandung and Surabaya. It was shown in the result that linear kernel function performs the best accuracy for the first data set with 0% validation error and 13% generalization error. While for the second data set, multi layer perceptron (MLP) kernel function has the best performance with 8% validation error and 37% generalization error. <br /> <p align="justify">