Application of Fourier Transform in Ultrasound Fatty Liver Imaging: Noise Reduction and Quantitative Analysis

傅立葉轉換在超音波脂肪肝影像處理中的應用：雜訊去除與定量分析

10.6717/JTMRT.202407_12(1).0002

Hsin-Yueh Su(蘇心悅)；Chi-Yuan Wang(王祺元)；Shang-Ting Hsieh(謝尚廷)；Wei-Chang Du(杜維昌)；Yung-Hui Huang(黃詠暉)

Fatty Liver ； Ultrasound ； K-Nearest-Neighbor Classifiers ； 脂肪肝 ； 超音波影像 ； KNN

臺灣醫事放射期刊

12卷1期（2024 / 07 / 01）

14 - 27

英文;繁體中文

Purpose: Incorporating fast Fourier transform into this study; the research aims to refine ultrasound imaging for fatty liver diagnosis. By applying fast Fourier transform techniques in image processing, the study seeks to enhance the denoising of ultrasound images, thereby improving image clarity and reducing the impacts of speckle noise. Methods and Materials: This retrospective study, approved by the institutional review board of Zuoying Branch of Kaohsiung Armed Forces General Hospital (Approval No: KAFGH 102- 013). For the practical aspect, B-mode images from ultrasound studies on 80 patients with abnormal liver function (AST, ALT levels over 40) were randomly selected. Meanwhile, 90 patients were selected with no finding. Image preprocessing utilized fast Fourier Transform (FFT) for noise reduction in images. Then, human identified and marked the areas of interest (ROI) in the images. Feature extraction and quantitative analysis was applied K-Nearest- Neighbor classifiers. The confusion matrix was used to evaluate the performance with specifically, sensitivity, accuracy, positive predicted value (PPV), negative predicted value (NPV), Kappa index, and area under ROC curve. Results: The optimal FFT thresholds, identified as 1.20 and 1.00, yielded the highest diagnostic performance as indicated by the AUC values of 0.828. These thresholds demonstrated improved accuracy, sensitivity, specificity, and predictive values, establishing FFT as a beneficial preprocessing step for ultrasound image analysis when tuned correctly. Conclusion: The use of FFT in preprocessing ultrasound images for the classification of liver conditions with KNN is validated, with the caveat that parameter optimization is crucial for its success. This study lays the groundwork for future research to refine these parameters and potentially combine FFT preprocessing with other advanced machine learning techniques to further enhance the diagnostic capabilities of ultrasound imaging in clinical practice.

目的：本研究納入快速傅立葉變換（FFT）主要目的是減少脂肪肝超音波影像雜訊，提高超音波影像品質從而提升影像定量與分類準確性。方法和材料：本回顧性研究已獲得左營海軍總醫院（Zuoying Branch of Kaohsiung Armed Forces General Hospital）機構審查委員會的批准（No：KAFGH 102-013）。隨機選取80名肝功能異常（AST、ALT水平超過40）的患者的B模式超音波像。同時，選取90名無異常發現（視為正常）的患者。影像預處理使用快速傅立葉變換（FFT）降低影像雜訊。然後，人工識別並標記了影像中的感興趣區域（ROI）進行特徵提取；定量分析應用統計96%信頼區間，採用KNN方法進行脂肪肝超音波影像分類。使用混淆矩陣評估KNN分類性能包括敏感性、準確性、陽性預測值（PPV）、陰性預測值（NPV）、Kappa指數和ROC曲線下面積。結果：FFT閾值的最佳選擇為1.20和1.00，具有的提分類高準確性；再者其ROC曲線下面積（AUC）值為0.828。正確調整閾值，能改進準確性、敏感性、特異性和預測值，FFT可作為脂肪肝超音波影像分析降噪的重要處理步驟。結論：KNN分類脂肪肝準確性從FFT處理超音波影像得到驗證，但需要注意的是參數優化對其成功至關重要。本研究為未來的研究奠定了基礎，將根據參數優化並可能將FFT預處理與其他先進的機器學習技術相結合，進一步提升臨床實踐中超音波影像的診斷能力。