保留三维体数据的mask的ROI区域---Python

针对最近的心脏细分割/肝脏分割项目，任务的划分分为两个阶段：

1、心脏/肝脏的粗定位；

2、心脏/肝脏的细分割；

第一步：粗定位（在原始标记上获取ROI）

def FindLabelBox3D(img, offset):
    '''
    img:ct-label-data
    offset:copy-level
    '''
    xdim = np.zeros(2)  # bouding box 和 x轴的交点
    ydim = np.zeros(2)  # bouding box 和 y轴的交点
    zdim = np.zeros(2)  # bouding box 和 z轴的交点
    tmp = np.squeeze(np.sum(np.sum(img, axis=2), axis=1))
    print("for x:", len(tmp))
    for i in range(len(tmp)):
        if tmp[i] == 0:
            xdim[0] = i
        else:
            break
    xdim[1] = len(tmp)
    for i in reversed(range(len(tmp))):
        if tmp[i] == 0:
            xdim[1] = i
        else:
            break
    # for y
    tmp = np.squeeze(np.sum(np.sum(img, axis=2), axis=0))
    print("for y:", len(tmp))
    for i in range(len(tmp)):
        if tmp[i] == 0:
            ydim[0] = i
        else:
            break

    ydim[1] = len(tmp)
    for i in reversed(range(len(tmp))):
        if tmp[i] == 0:
            ydim[1] = i
        else:
            break
    # for z
    tmp = np.squeeze(np.sum(np.sum(img, axis=1), axis=0))
    print("for z:", len(tmp))
    for i in range(len(tmp)):
        if tmp[i] == 0:
            zdim[0] = i
        else:
            break

    zdim[1] = len(tmp)
    for i in reversed(range(len(tmp))):
        if tmp[i] == 0:
            zdim[1] = i
        else:
            break

    # offset
    xdim[0] = max(0, xdim[0] - offset)
    xdim[1] = min(np.size(img, 0), xdim[1] + offset)

    ydim[0] = max(0, ydim[0] - offset)
    ydim[1] = min(np.size(img, 1), ydim[1] + offset)

    zdim[0] = max(0, zdim[0] - offset)
    zdim[1] = min(np.size(img, 2), zdim[1] + offset)

    return xdim, ydim, zdim

上述函数可以根据label-标记获取到ROI区域；

获取对应的label-区域与image-区域：

xdim, ydim, zdim = FindLabelBox3D(label, 0)
##image-roi
image_bouding_box = image_data[xdim[0]:xdim[1], ydim[0]:ydim[1], zdim[0]:zdim[1]]
##label-roi
label_bouding_box = label_data[xdim[0]:xdim[1], ydim[0]:ydim[1], zdim[0]:zdim[1]]

上述的image_bouding_box与label_bouding_box即为获取的ROI。