| Medical imaging has historically been very successful in exposing a patient's anatomy beyond external visibility, thus allowing more efficient and accurate treatments. The field of medicine continues to search for new techniques in order to increase accuracy, reduce complications, enable real-time feedback, allow early detection, and reduce human errors. Various medical imaging techniques exist, but suffer from multiple correlated noise or performance that hinders real-time application, i.e. ultrasounds. We deal with the real-time image segmentation in B-Scan ultrasounds using machine learning for template design, parallelized Hybrid Median filter for speckle removal, and Cellular Neural Network for image processing, and then analyze the B-scan image features. Our contribution attempts to detect edges efficiently and enable real-time procedures. We implemented a smart CNN system for edge detection on ASIC 45nm technology in order to be able to exploit parallelism and real-time capabilities. We were able to get the desired visual results at 100MHz frequency for every 38x38 blocks with 182.20s completion time for full 512x512 ultrasound image to be fully processed, which is enough for 70fps real-time ultrasound applications. |
