在使用LIDC資料集做肺結節檢測時對肺區進行分割後效果會好很多，主要是減少肺區以外的組織對檢測的干擾。

方法比較簡單：

def img_segmentation(dicom_file):
    threshold = -480 / dicom_file.RescaleSlope - dicom_file.RescaleIntercept
    mask=dicom_file.pixel_array
    mask=(mask>threshold).astype(np.float32)
    element = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (10, 10))
    mask = cv2.erode(mask, element)
    cols,rows=mask.shape[:2]
    #cv2.imshow('mask', mask)
    #cv2.waitKey()
    mask_back=mask.copy()
    flood_mask=np.zeros([cols+2,rows+2],dtype=np.uint8)
    cv2.floodFill(mask,flood_mask,(0,0),255)
    element = cv2.getStructuringElement(cv2.MORPH_ELLIPSE, (4, 4))
    mask = cv2.dilate(mask, element)
    #cv2.imshow('mask', mask)
    #cv2.waitKey()
    #mask=(mask.astype(np.uint8))*255
    #mask=255-mask
    #return mask
    #seg=mask*(dicom_file.pixel_array)
    mask=1.0-mask/255
    mask=mask-mask_back
    mask=mask*(mask>0)
    #mask=(mask.astype(np.uint8))*
    print mask.max(), mask.min()
    flood_mask = np.zeros([cols + 2, rows + 2], dtype=np.uint8)
    cv2.floodFill(mask, flood_mask, (0, 0), 255)
    print mask.max(), mask.min()
    mask=mask.astype(np.uint8)
    mask=(mask<128).astype(np.float32)
    #cv2.imshow('mask',mask)
    #cv2.waitKey()
    return mask