Feature Analysis Of X-ray Chest Radiographs

With the computer and the rapid development of imaging technology, CAD(Computer-aided diagnosis, CAD) has become a research hotspot. It has been proven to be avery effective approach as assistant to radiologists for improving diagnostic accuracy byproviding a"second opinion". The most commonly used application of CAD for X-ray is thedetection of lung cancer. Radiologists analyze all kinds of lung disease by Chest X-ray images.The doctor is concerned at the lung area which is considered as the interesting region.Therefore, the first step of chest X-ray image analysis is to mark the outline of lung region sothat the computer can do the lesion recognition processing for the next step, and provide areliable diagnostic data foundation for doctors.Firstly, this paper describes the digital chest X-ray medical image pre-processing andanalysis technique, including histogram equalization, high-frequency mixing and multi-scaleRetinex algorithm. On this basis, the commonly used medical image segmentation arecompared and summarized, and then applied to the lung segmentation according to actualsituation. By comparing the typical segmentation method, including the threshold,morphology, edge detection, and model method based on parameters and so on, it analyzestheir advantages and disadvantages. Because of the fuzzy boundaries between the varioustissues in human body, there are some difficulties in the accurate segmentation of lung. Thispaper presents an interactive lung segmentation method. First, using optimized Gabor oddfilter to filter chest X-ray image, so we can get a map of edge response energy. Then, usingthe edge response energy values to construct the live-wire cost function, which cansegmentation the chest image. Test shows that this algorithm can correctly distinguish strongand weak edge, and also can extract the lung's contour quickly and effectively. Comparedwith the traditional algorithm can reduce the number of human-computer interaction, morerobustness and efficiency advantages. As we know, lung nodule in Chest X-ray image showssimilar to the circular structure whose density is higher than the surrounding and it has variable length and diameter. On this point, this paper discusses lung nodule detectionalgorithm at last. Also, it uses multi-scale lung nodule enhancement and multi-scalebackground suppression to obtain the candidate lung nodule region, puts forward an improvedcandidate lung nodule detection method - the difference image technique. Experiments showthat this method, which is better than traditional methods when getting pulmonary nodulecandidate region, can reduce some false positives.