Development of Image Processing Schemes for Improving Needle Visibility in Ultrasonography (USG) for Medical Needle Guided

• Main Author: Septyvergy, Arkanty
• Format: Theses
• Language: Indonesia
• Subjects: Teknik (Rekayasa, enjinering dan kegiatan berkaitan)
• Online Access: https://digilib.itb.ac.id/gdl/view/45085
• Institution: Institut Teknologi Bandung

Minimally invasive interventional procedures are procedures that are often used in the medical field. This is because this procedure only needed a small incision to insert a medical tool so it can minimize injuries and speed up healing of postoperative patients. The tool that often used is needles in the anesthesia procedure. According to a survey conducted in a Turkish private hospital the epidural anesthesia procedure increase from 57% to 95%. However, there was a failure of 0,50% from the total implementation of the procedure. Meanwhile, based on a survey conducted in the United Kingdom, complications caused by inaccurate implementation of epidural anesthesia procedures were reported at 0,19% to 3,60%. One way to reduce the failure of the implementation of anesthetic procedures and complications is to use ultrasound (USG) images. This image is used to visualize the position of the needle so it can guide the doctors in carrying out epidural anesthesia procedures. But the use of ultrasound has disadvantage such as inconsistent needle visibility. This is happening because the typical noise that ultrasound has and the acoustic phenomenon between the needle and ultrasonic waves that occurs so the position of the needle is difficult to detect. There have been several efforts to increase needle visibility, such as manipulating beam steering and the interaction of ultrasonic waves with needles. But when these efforts have been made, it turns out the needle's visibility is still not consistent and the position of the needle is still difficult to detect. Based on that, the researchers tried to make another approach by doing post processing on the ultrasound image. Post processing that is commonly done is by detecting needle position using hough transformation or RANSAC algorithm. Hough transformation method has some lack which is required binaryization or edge detection. Binaryzation or edge detection requires a thresholding value that depends on image conditions. In the RANSAC algorithm, if the iterations performed are not appropriate, the resulting solution is not optimal. Based on that lack, in this reseach a scheme for image processing was developed to improve needle visibility. The proposed scheme starts with selecting the Region of Interest (ROI) of the area around the needle in the ultrasound image. Stab angle iv of needle will be estimated from ROI. The estimated stab angle will be used to rotate the image so the needle image becomes horizontal. The estimated position of the needle will be calculated using a linear derivative in each pixel. Needle position estimation using linear derivatives still produces outlier data, so the outlier data removing method is needed. The outlier removing uses two methods, namely a combination of the median moving method and the moving median absolute deviation and the outlier removing method using a mean and standard deviation. The resulting data then carried out by polynomial interpolation so that it can detect the needle position perfectly. Needle tip detection is carried out to limit the end of the needle tip detection. Schemes that have been built are then tested in three image groups. The first data is needle stabbing data that is driven by a robot and its image is taken using a digital camera. The needle movement is varied by the depth and angle of needle puncture. The first data is tested to ensure that the scheme that has been created actually detects needle parts. The second data is the first data added by speckle noise to represent the image produced by USG. The third data is an image of a needle stabbing sequence that stabbing in a water medium and captured its image using ultrasound. Each sequence has 3 needle stab images with different depths. The scheme that has been created can detect the position of the needle with a success of 96% of the first data and the second data. In the third data, the scheme can detect needle position by 90%. The failure is caused by large noise generated from the reflection of the transducer support or needle support.