Feature Enhancement of Medical Ultrasound Scans Using Multifractal Measures

In the field of ultrasound image analysis, speckle artefact removal algorithms play an important role in improving the diagnostic value of the images. It is well known that speckle filtering algorithms generally induce blurring and other unwanted artefacts. It is therefore often required to have both speckle reduction and feature improvement to enhance the overall visual quality of medical ultrasound video frames. In this paper, we use a convolutional neural network for effective speckle reduction. We also propose a novel method based on multifractal measures to characterize texture features and use them to enhance the contrast, and also certain types of resolvable details, and structure in ultrasound video frames. Four different types of multifractal measures are used for comparative analysis. Ultrasound scans of multiple anatomical structures are used for experimental validation of the proposed methods. Subjective and objective evaluation by clinical experts and three quality metrics indicate that the proposed pipeline with the inverse-min multifractal measure is able to reduce speckle content effectively while preserving visual quality, information content, and improving features useful for better clinical interpretation.