Application of two-dimensional fractional-order convolution and bounding box pixel analysis for rapid screening of pleural effusion.

Pleural effusion is a pathologic symptom in which there is accumulation of body fluids around the lungs. A chest radiograph is a rapid examination technique and does not require complex setup for making a preliminary diagnosis of lung and heart diseases. In radiographic visualization, the symptom patterns appear as light or dark areas in the lung cavity. Computer-aided diagnosis is an automatic manner that can rapidly highlight the object region by preanalyzing medical images. It can improve the problems of manual inspection and allow diagnosis in remote medical facilities. Based on the ratios of lung anatomy, the automatic screening manner based on pattern recognition can be viewed as pixel value detection in the bilateral lung cavities. In this study, a fractional-order convolution (FOC) process is used to enhance the original image for an accurate extrapolation of the desired object in an image. The specific object image feature can be improved, and an accurate quantification of the pleural effusion region can be obtained using the suitable ranges of fractional-order parameters. Based on the boundaries of homogeneous regions, the pixel ratios of the lung anatomy between normal and abnormal conditions can be computed. The pleural effusion sizes and volumes can be rapidly estimated through the number of pixel changes. The experimental results reveal that the feature maps are similar and stable on image enhancement and segmentation with two fractional-order enhancement masks, as fractional-order v = 0.05 to 0.20 for mask 1# and v = 0.80 to 0.95 for mask 2#, respectively. The results also demonstrate the feasibility of the study on combining two-dimensional image FOC-process and bounding box pixel analysis to estimate the moderate and large effusion sizes from 500-2,000 mL.