An Accurate and Robust Method for Detecting Fetal Heads in Ultrasound Images Based on Integrating a Voting Scheme and an Improved IRHT

摘要

To achieve accurate and robust detection of fetal heads in US (ultrasound) images, this paper proposes a method that integrates a voting scheme and an improved iterative randomized Hough transform (IRHT) method. First, a skeleton image is extracted from the input US image by pre-processes. Next, the voting scheme, which applies the improved IRHT method to detect ellipses from the skeleton image, is repeated P times, in each of which, if the difference between each detected ellipse’s parameter values and each parameter values saved in the list is small, the accumulator of the nearest values in the list is incremented by one (voting); otherwise, the detected p arameter values are appended to the list. Finally, the ellipse with the most voted values in the list is outputted as the final detection result. As a result of experiments that use 30 US images captured under different conditions, it turns out that the proposed method achieves a very high ellipse detection rate of 94.97%, which is much higher than IRHT and improved IRHT. In addition, some parameters concerning the proposed method such as the optimal value for P are experimentally explored.