Computer aided diagnosis system for lung cancer CT images

本文报道了用于肺癌CT计算机辅助诊断的图像处理技术。LSCT是我们项目组新开发的用于肺癌大规模筛查的移动式CT扫描仪。在这种新的LSCT系统中，一个关键问题是将1张x射线胶片的图像信息增加到每人30张左右。我们试图通过以下图像处理技术，大幅减少要显示给医生的图像信息。1 approach （MIP method）考虑MIP method（最大强度投影），将患者肺组织30片切片组成的三维信息通过投影显示在二维平面上。将MIP方法应用于LSCT的最重要问题是预先删除可能掩盖病理阴影的干扰器官信号。我们利用数学形态学、Split-Quoit滤波、模型信息和活动轮廓模型（snake）开发了一种新的稳定的方法。该方法应用于68例患者样本，结果令人满意。提出了一种自动识别病理阴影候选区域的方法。通过只向医生显示包含候选区域的切片，可以大大减少要诊断的横截面的数量。肺癌的阴影存在于肺野，呈孤立状，面积小。利用这一特性，开发了2-D、3-D Quoit滤波器（Q-filters）和MIP-2D-Q滤波器来自动提取孤立阴影，其中q滤波器是基于数学形态学的滤波器。该方法应用于68个患者样本，包括1809张图像，将显示给医生诊断的图像减少到144张，即减少率为8%（2.1张/患者）。其中5例肿瘤患者被正确提取。

This paper reports the image processing technique for computer-aided diagnosis of lung cancer by CT (LSCT). LSCT is the newly developed mobile-type CT scanner for the mass screening of lung cancer by our project team. In this new LSCT system, one essential problem is the increase of image informantion to about 30 slices per person from 1 X-ray film. We tried to reduce the image information drastically to be displayd for the doctor, by image processing techniques, as follows.NO.1 approach (MIP method)MIP mehtod (maximum intensity projection)is considered, where the three-dimensional information composed of 30 slices of patient's lung tissue is displayd by a projection on the two-dimensional plane. The most important problem to apply MIP method for LSCT is to delete beforehand the interfering organ signals that could mask the pathological shadows. We developed a new and stable method for it using mathematical morphology, Split-Quoit filter, model information, and active contour model (snake).This method is applied for 68 patient samples and the result is promissing.NO.2 approach (Quoit filter method)A new method is introduced which automatically recognizes the candidate regions for the pathological shadows. By displaying only the slices containing the candidates regions to the doctor, the number of cross sections to bediagnosed can be drastically reduced.The shadow of the lung cancer which exists in the lung field, appears isolated and with a small area. By utilizing this property, 2-D , 3-D Quoit filters (Q-filters) and MIP-2D-Q filter are developed to extract the isolated shadow automatically, where Q-filters are those based on mathematical morphology.This method is applied for 68 patient samples including 1809 images, and the number of displayd images to be diagnosed by the doctor is reduced to 144 images, which means the reductin rate 8% (2.1 images/patient). Among those, the 5 cancer patients are extracted correctly.