Ultrasound Image Segmentation for Deep Vein Thrombosis using Unet-CNN based on Denoising Filter

Deep vein thrombosis (DVT) is caused by an abnormal blood clot condition in the network of blood vessels. Several risk factors that often cause DVT are advanced age, post-surgery, hospitalization, pregnant women, and obesity. In general, Diagnosis of DVT uses ultrasound images. However, diagnosis using ultrasound manually takes a long time, and the accuracy of image reading depends on medical personnel. It requires a system that can detect DVT automatically. Also, it can be obtained quickly and has good accuracy. This study proposes a segmentation model for ultrasound images of deep vein thrombosis using U-Net CNN based on a denoising filter. Furthermore, calculating the suspected area to be DVT predicted using U -Net CNN. The denoising filter consisted of eight filters. That model system was tested with an ultrasound image dataset. The dataset was obtained from four volunteers. The volunteers have been identified as having symptoms of deep vein thrombosis. The dataset was captured and recorded using an ultrasound device carried out by medical experts. Each DVT recorded dataset is extracted into frames. The full frames obtained are 317 frames. Then the ultrasound image data is manually labeled by medical personnel. The experimental results show that the Gaussian filter has the best results, with 99% of accuracy and 0.0252 scores of an average loss parameter. Meanwhile, the DVT prediction test using U-Net CNN segmentation based on the calculation of the mean IoU is 84.9% accurate. The measure of the mean Hausdorff distance is 4.17 of the score. We want to investigate the detection and classification of DVT for further research.