High scattering sensitivity entropy imaging for breast tumor characterization and classification

Authors:
Publisher website
Google Scholar
Abstract
Background: Diagnosing and characterizing breast lesions and tumors remains a common challenge in clinical practice. Ultrasound imaging stands out for its safety, real‐time capability, and affordability. However, the image quality of conventional ultrasound examination is limited, and the diagnosis of ultrasonographic images depends heavily on the experience of the sonographer. Therefore, improving ultrasound images and extracting tissue information from ultrasound signals to provide auxiliary means is crucial for accurate breast tumor diagnosis. Purpose: Medical ultrasound imaging has been widely used in clinical diagnosis. However, traditional ultrasound has limitations in the diagnosis of breast soft tissue diseases. This study proposed a high scattering sensitivity fuzzy entropy (FE) imaging method to enhance image contrast and improve detectability for breast tumors. Moreover, this imaging method can make a preliminary classification and characterization of benign and malignant breast lesions through quantitative analysis of ultrasound radio frequency data and the calculation of the entropy value without biopsy examination. Methods: To achieve the fuzzy entropy imaging, a sliding window is selected to traverse across the image with a step of one sampling point while the entropy value within the sliding window is calculated. This entropy value is assigned to the center pixel of the window. The parametric image was obtained after the entropy values of all pixels were calculated. During the clinical experiments, the breast lesions were classified as benign or malignant by biopsy examination. After entropy imaging, the average entropy value of the lesion area was calculated. The entropy values of all cases of benign and malignant tumors were averaged, respectively, to verify whether the fuzzy entropy can characterize the breast lesions. All the statistical analysis was conducted by one‐sample t‐test to obtain the mean value and standard deviation. The Tukey test was performed, and the effect size of Cohen's d was calculated to verify whether there was a significant difference between the entropy value of benign lesions and malignant lesions. Results: In the clinical breast imaging experiment, the FE method obtained the highest Matthews correlation coefficient (MCC) of 0.875 ± 0.047 ( p < 0.0001) and F1 score of 0.876 ± 0.049 ( p < 0.0001). The MCC and F1 scores of FE imaging were significantly different from those of other entropy imaging methods in the Tukey test ( p < 0.0001). The effect sizes of Cohen's d of F1 score of FE method compared with the WSE method and hNSE method were 1.498 and 1.107, respectively. The contrast‐to‐noise ratio (CNR) of FE images increased by 124.37% ( p < 0.0001) compared with B‐mode images (5.210 ± 3.136, p < 0.0001). The above results show that the FE method has good comprehensive performance in improving the detection accuracy and contrast of breast lesions. The fuzzy entropy value of benign tumors (0.033 ± 0.0.14, p < 0.0001) is higher than that of malignant tumors (0.022 ± 0.013, p < 0.0001) with both statistical and practical significance, indicating that the benign and malignant tumors can be characterized and classified by fuzzy entropy value. Conclusions: The proposed ultrasound fuzzy entropy breast imaging method can effectively improve the ultrasound imaging performance and the ability to detect lesions, because fuzzy entropy can measure the microscopic chaos of breast tissue and enhance the scattering information characteristics in the signal. Meanwhile, fuzzy entropy imaging can classify benign and malignant lesions, because fuzzy entropy considers the causality within the ultrasound signal, avoiding information aliasing and loss, so that it can detect weaker information in the signal and can reflect organizational information more accurately.
Funding Information
  • National Natural Science Foundation of China (21927803, 82402279)
  • National Key Research and Development Program of China (2022YFF0706500)

This publication has 37 references indexed in Scilit: