Enhancing Semantic Segmentation in Chest X-Ray Images through Image Preprocessing: ps-KDE for Pixel-wise Substitution by Kernel Density Estimation

• DOI: 10.1101/2024.02.15.24302871
• Publication Date: 2024-02-17T23:04:14Z
• Authors: Yuanchen Wang; Yujie Guo; Ziqi Wang; Linzi Yu; Yujie Yan; Zifan Gu
• Keywords: Substitution (logic); Kernel density estimation; Kernel (algebra)

Background Deep-learning-based semantic segmentation algorithms, in combination with image preprocessing techniques, can reduce the need for human annotation and advance disease classification. Among established CLAHE has demonstrated efficacy enhancing segmentations algorithms across various modalities. Method This study proposes a novel technique, ps-KDE, to investigate its impact on deep learning segment major organs posterior-anterior chest X-rays. Ps-KDE augments contrast by substituting pixel values based their normalized frequency all images. Our approach employs U-Net architecture ResNet34 (pre-trained ImageNet) serving as decoder. Five separate models are trained heart, left lung, right clavicle, clavicle. Results The model lung using ps-KDE achieved Dice score of 0.780 (SD=0.13), while that 0.717 (SD=0.19), p <0.01. also appears be more robust CLAHE-based misclassified lungs select test images model. Discussion results suggest offers advantages over current techniques when segmenting certain regions. could beneficial subsequent analysis such classification risk stratification.