Convolutional Neural Network untuk Pendeteksian Patah Tulang Femur pada Citra Ultrasonik B–Mode

Rika Rokhana (Institut Teknologi Sepuluh Nopember)
Joko Priambodo (Institut Teknologi Sepuluh Nopember)
Tita Karlita (Institut Teknologi Sepuluh Nopember)
I Made Gede Sunarya (Institut Teknologi Sepuluh Nopember)
Eko Mulyanto Yuniarno (Institut Teknologi Sepuluh Nopember)
I Ketut Eddy Purnama (Institut Teknologi Sepuluh Nopember)
Mauridhi Hery Purnomo (Institut Teknologi Sepuluh Nopember)

Jurnal Nasional Teknik Elektro dan Teknologi Informasi
Vol 8 No 1: Februari 2019

The bone fracture detection using X–rays or CT–scan produces accurate images but has harmful effect radiation. This paper presented the use of ultrasonic waves (US) as an alternative to substitute those two instruments. This study used femur bovine and chicken bones in conditions with and without meat. The fractures are artificially made on transverse and oblique patterns. The scanning US probe produces two-dimensional (2D) B–mode images. Fracture detection is done using five variations of the Convolutional Neural Network (CNN) architectural design, i.e., CNN1–CNN5. The results showed that the CNN4 is the best design of bone contour recognition and bone fracture classification compared to the other tested designs, with 95.3% accuracy, 95% sensitivity, and 96% specificity. The comparison with the Support Vector Machine (SVM) and k-NN classification methods indicate that CNN has superior performance in accuracy, sensitivity, and specificity.