Application and Optimization of Lee Filter for Segmentation of Benign Tumor in Breast Ultrasound Images.

Aims: In this study, we aimed to optimize the window size of Lee filter for speckle noise reduction in breast ultrasound images (BUSIs) by evaluating the segmentation performance of benign tumors using a U-Net model. Subjects and Methods: Benign tumor images were acquired with added speckle noise (intensity = 0.05) to obtain the noisy images. A Lee filter was applied to noisy images with window sizes of 3 × 3, 5 × 5, 7 × 7, and 9 × 9. To evaluate noise reduction and segmentation performance, noisy and filtered images were used as input images for the U-Net model. The segmentation performance according to various window sizes of the Lee filter was quantitatively evaluated using intersection over union (IoU), peak signal to noise ratio (PSNR), and universal quality image index (UQI). Results: As a result, a window size of 7 × 7 achieved the highest performance across all evaluation factors. In particular, when comparing the image with a 7 × 7 window size to the noisy image, the segmentation performance exhibited average improvements of approximately 19%, 5%, and 19% in IoU, PSNR, and UQI, respectively, with maximum improvements of approximately 27%, 7%, and 23%. Conclusions: This study demonstrated that applying the Lee filter with an optimized 7 × 7 window size improved speckle noise reduction and segmentation accuracy in BUSIs, supporting the applicability of deep learning-based tumor segmentation in clinical practice.