tools.py

__author__ = 'Ryba'
import numpy as np
import matplotlib.pyplot as plt
import skimage.exposure as skiexp
from skimage.segmentation import mark_boundaries
import os
import glob
import dicom
import cv2
# from skimage import measure
import skimage.measure as skimea
import skimage.morphology as skimor
import skimage.transform as skitra
import skimage.filter as skifil
import skimage.restoration as skires
import skimage.filter as skifil
import skimage.segmentation as skiseg
import scipy.stats as scista
import scipy.ndimage.morphology as scindimor
import scipy.ndimage.measurements as scindimea
import scipy.ndimage.interpolation as scindiint

import pickle

#import py3DSeedEditor

#----------------------------------------------------------------------------------------------------------------------
#----------------------------------------------------------------------------------------------------------------------
def get_seeds(im, minT=0.95, maxT=1.05, minInt=0, maxInt=255, debug=False):
    vals = im[np.where(np.logical_and(im>=minInt, im<=maxInt))]
    hist, bins = skiexp.histogram(vals)
    max_peakIdx = hist.argmax()

    minT *= bins[max_peakIdx]
    maxT *= bins[max_peakIdx]
    histTIdxs = (bins >= minT) * (bins <= maxT)
    histTIdxs = np.nonzero(histTIdxs)[0]
    class1TMin = minT
    class1TMax = maxT

    seed_mask = np.where( (im >= class1TMin) * (im <= class1TMax), 1, 0)

    if debug:
        plt.figure()
        plt.plot(bins, hist)
        plt.hold(True)

        plt.plot(bins[max_peakIdx], hist[max_peakIdx], 'ro')
        plt.plot(bins[histTIdxs], hist[histTIdxs], 'r')
        plt.plot(bins[histTIdxs[0]], hist[histTIdxs[0]], 'rx')
        plt.plot(bins[histTIdxs[-1]], hist[histTIdxs[-1]], 'rx')
        plt.title('Image histogram and its class1 = maximal peak (red dot) +/- minT/maxT % of its density (red lines).')
        plt.show()

    #minT *= hist[max_peakIdx]
    #maxT *= hist[max_peakIdx]
    #histTIdxs = (hist >= minT) * (hist <= maxT)
    #histTIdxs = np.nonzero(histTIdxs)[0]
    #histTIdxs = histTIdxs.astype(np.int)minT *= hist[max_peakIdx]
    #class1TMin = bins[histTIdxs[0]]
    #class1TMax = bins[histTIdxs[-1]

    #if debug:
    #    plt.figure()
    #    plt.plot(bins, hist)
    #    plt.hold(True)
    #
    #    plt.plot(bins[max_peakIdx], hist[max_peakIdx], 'ro')
    #    plt.plot(bins[histTIdxs], hist[histTIdxs], 'r')
    #    plt.plot(bins[histTIdxs[0]], hist[histTIdxs[0]], 'rx')
    #    plt.plot(bins[histTIdxs[-1]], hist[histTIdxs[-1]], 'rx')
    #    plt.title('Image histogram and its class1 = maximal peak (red dot) +/- minT/maxT % of its density (red lines).')
    #    plt.show()

    return seed_mask, class1TMin, class1TMax


#----------------------------------------------------------------------------------------------------------------------
#----------------------------------------------------------------------------------------------------------------------
def seeds2superpixels(seed_mask, superpixels, debug=False, im=None):
    seeds = np.argwhere(seed_mask)
    superseeds = np.zeros_like(seed_mask)

    for s in seeds:
        label = superpixels[s[0], s[1]]
        superseeds = np.where(superpixels==label, 1, superseeds)

    if debug:
        plt.figure(), plt.gray()
        plt.subplot(121), plt.imshow(im), plt.hold(True), plt.plot(seeds[:,1], seeds[:,0], 'ro'), plt.axis('image')
        plt.subplot(122), plt.imshow(im), plt.hold(True), plt.plot(seeds[:,1], seeds[:,0], 'ro'),
        plt.imshow(mark_boundaries(im, superseeds, color=(1,0,0))), plt.axis('image')
        plt.show()

    return superseeds


#----------------------------------------------------------------------------------------------------------------------
#----------------------------------------------------------------------------------------------------------------------
def intensity_range2superpixels(im, superpixels, intMinT=0.95, intMaxT=1.05, debug=False, intMin=0, intMax=255):#, fromInt=0, toInt=255):

    superseeds = np.zeros_like(superpixels)

    #if not intMin and not intMax:
    #    hist, bins = skexp.histogram(im)
    #
    #    #zeroing values that are lower/higher than fromInt/toInt
    #    toLow = np.where(bins < fromInt)
    #    hist[toLow] = 0
    #    toHigh = np.where(bins > toInt)
    #    hist[toHigh] = 0
    #
    #    max_peakIdx = hist.argmax()
    #    intMin = intMinT * bins[max_peakIdx]
    #    intMax = intMaxT * bins[max_peakIdx]

    sp_means = np.zeros(superpixels.max()+1)
    for sp in range(superpixels.max()+1):
        values = im[np.where(superpixels==sp)]
        mean = np.mean(values)
        sp_means[sp] = mean

    idxs = np.argwhere(np.logical_and(sp_means>=intMin, sp_means<=intMax))
    for i in idxs:
        superseeds = np.where(superpixels==i[0], 1, superseeds)

    if debug:
        plt.figure(), plt.gray()
        plt.imshow(im), plt.hold(True), plt.imshow(mark_boundaries(im, superseeds, color=(1,0,0)))
        plt.axis('image')
        plt.show()

    return superseeds


def show_slice(data, segmentation=None, lesions=None, show='True'):
    plt.figure()
    plt.gray()
    plt.imshow(data)

    if segmentation is not None:
        plt.hold(True)
        contours = skimea.find_contours(segmentation, 1)
        for contour in contours:
            plt.plot(contour[:, 1], contour[:, 0], 'b', linewidth=2)

    if lesions is not None:
        plt.hold(True)
        contours = skimea.find_contours(lesions, 1)
        for contour in contours:
            plt.plot(contour[:, 1], contour[:, 0], 'r', linewidth=2)

    plt.axis('image')

    if show:
        plt.show()


def change_slice_index(data):
    n_slices = data.shape[2]
    data_reshaped = np.zeros(np.hstack((data.shape[2], data.shape[0], data.shape[1])))
    for i in range(n_slices):
        data_reshaped[i, :, :] = data[:, :, i]
    return data_reshaped


def read_data(dcmdir, indices=None, wildcard='*.dcm', type=np.int16):

    dcmlist = []
    for infile in glob.glob(os.path.join(dcmdir, wildcard)):
        dcmlist.append(infile)

    if indices == None:
        indices = range(len(dcmlist))

    data3d = []
    for i in range(len(indices)):
        ind = indices[i]
        onefile = dcmlist[ind]
        if wildcard == '*.dcm':
            data = dicom.read_file(onefile)
            data2d = data.pixel_array
            try:
                data2d = (np.float(data.RescaleSlope) * data2d) + np.float(data.RescaleIntercept)
            except:
                print('problem with RescaleSlope and RescaleIntercept')
        else:
            data2d = cv2.imread(onefile, 0)

        if len(data3d) == 0:
            shp2 = data2d.shape
            data3d = np.zeros([shp2[0], shp2[1], len(indices)], dtype=type)

        data3d[:,:,i] = data2d

    #need to reshape data to have slice index (ndim==3)
    if data3d.ndim == 2:
        data3d.resize(np.hstack((data3d.shape,1)))

    return data3d


def windowing(data, level=50, width=300, sub1024=False, sliceId=2):
    #srovnani na standardni skalu = odecteni 1024HU
    if sub1024:
        data -= 1024

    #zjisteni minimalni a maximalni density
    minHU = level - width
    maxHU = level + width

    if data.ndim == 3:
        if sliceId == 2:
            for idx in range(data.shape[2]):
                #rescalovani intenzity tak, aby skala <minHU, maxHU> odpovidala intervalu <0,255>
                data[:, :, idx] = skiexp.rescale_intensity(data[:, :, idx], in_range=(minHU, maxHU), out_range=(0, 255))
        elif sliceId == 0:
            for idx in range(data.shape[0]):
                #rescalovani intenzity tak, aby skala <minHU, maxHU> odpovidala intervalu <0,255>
                data[idx, :, :] = skiexp.rescale_intensity(data[idx, :, :], in_range=(minHU, maxHU), out_range=(0, 255))
    else:
        data = skiexp.rescale_intensity(data, in_range=(minHU, maxHU), out_range=(0, 255))

    return data.astype(np.uint8)


def smoothing(data, d=10, sigmaColor=10, sigmaSpace=10, sliceId=2):
    if data.ndim == 3:
        if sliceId == 2:
            for idx in range(data.shape[2]):
                data[:, :, idx] = cv2.bilateralFilter( data[:, :, idx], d=d, sigmaColor=sigmaColor, sigmaSpace=sigmaSpace )
        elif sliceId == 0:
            for idx in range(data.shape[0]):
                data[idx, :, :] = cv2.bilateralFilter( data[idx, :, :], d=d, sigmaColor=sigmaColor, sigmaSpace=sigmaSpace )
    else:
        data = cv2.bilateralFilter( data, d=d, sigmaColor=sigmaColor, sigmaSpace=sigmaSpace )
    return data


def smoothing_bilateral(data, sigma_space=15, sigma_color=0.05, pseudo_3D='True', sliceId=2):
    if data.ndim == 3 and pseudo_3D:
        if sliceId == 2:
            for idx in range(data.shape[2]):
                # temp = skifil.denoise_bilateral(data[:, :, idx], sigma_range=sigma_color, sigma_spatial=sigma_space)
                temp = skires.denoise_bilateral(data[:, :, idx], sigma_range=sigma_color, sigma_spatial=sigma_space)
                data[:, :, idx] = (255 * temp).astype(np.uint8)
        elif sliceId == 0:
            for idx in range(data.shape[0]):
                # temp = skifil.denoise_bilateral(data[idx, :, :], sigma_range=sigma_color, sigma_spatial=sigma_space)
                temp = skires.denoise_bilateral(data[idx, :, :], sigma_range=sigma_color, sigma_spatial=sigma_space)
                data[idx, :, :] = (255 * temp).astype(np.uint8)
    else:
        # data = skifil.denoise_bilateral(data, sigma_range=sigma_color, sigma_spatial=sigma_space)
        data = skires.denoise_bilateral(data, sigma_range=sigma_color, sigma_spatial=sigma_space)
        data = (255 * data).astype(np.uint8)
    return data


def smoothing_tv(data, weight=0.1, pseudo_3D=True, multichannel=False, sliceId=2):
    if data.ndim == 3 and pseudo_3D:
        if sliceId == 2:
            for idx in range(data.shape[2]):
                # temp = skifil.denoise_tv_chambolle(data[:, :, idx], weight=weight, multichannel=multichannel)
                temp = skires.denoise_tv_chambolle(data[:, :, idx], weight=weight, multichannel=multichannel)
                data[:, :, idx] = (255 * temp).astype(np.uint8)
        elif sliceId == 0:
            for idx in range(data.shape[0]):
                # temp = skifil.denoise_tv_chambolle(data[idx, :, :], weight=weight, multichannel=multichannel)
                temp = skires.denoise_tv_chambolle(data[idx, :, :], weight=weight, multichannel=multichannel)
                data[idx, :, :] = (255 * temp).astype(np.uint8)
    else:
        # data = skifil.denoise_tv_chambolle(data, weight=weight, multichannel=False)
        data = skires.denoise_tv_chambolle(data, weight=weight, multichannel=False)
        data = (255 * data).astype(np.uint8)
    return data


def smoothing_gauss(data, sigma=1, pseudo_3D='True', sliceId=2):
    if data.ndim == 3 and pseudo_3D:
        if sliceId == 2:
            for idx in range(data.shape[2]):
                temp = skifil.gaussian_filter(data[:, :, idx], sigma=sigma)
                data[:, :, idx] = (255 * temp).astype(np.uint8)
        elif sliceId == 0:
            for idx in range(data.shape[0]):
                temp = skifil.gaussian_filter(data[idx, :, :], sigma=sigma)
                data[idx, :, :] = (255 * temp).astype(np.uint8)
    else:
        data = skifil.gaussian_filter(data, sigma=sigma)
        data = (255 * data).astype(np.uint8)
    return data


def analyse_histogram(data, roi=None, debug=False, dens_min=20, dens_max=255, minT=0.95, maxT=1.05):
    if roi == None:
        #roi = np.ones(data.shape, dtype=np.bool)
        roi = np.logical_and(data >= dens_min, data <= dens_max)

    voxels = data[np.nonzero(roi)]
    hist, bins = skiexp.histogram(voxels)
    max_peakIdx = hist.argmax()

    minT = minT * hist[max_peakIdx]
    maxT = maxT * hist[max_peakIdx]
    histTIdxs = (hist >= minT) * (hist <= maxT)
    histTIdxs = np.nonzero(histTIdxs)[0]
    histTIdxs = histTIdxs.astype(np.int)

    class1TMin = bins[histTIdxs[0]]
    class1TMax = bins[histTIdxs[-1]]

    liver = data * (roi > 0)
    class1 = np.where( (liver >= class1TMin) * (liver <= class1TMax), 1, 0)

    if debug:
        plt.figure()
        plt.plot(bins, hist)
        plt.hold(True)

        plt.plot(bins[max_peakIdx], hist[max_peakIdx], 'ro')
        plt.plot(bins[histTIdxs], hist[histTIdxs], 'r')
        plt.plot(bins[histTIdxs[0]], hist[histTIdxs[0]], 'rx')
        plt.plot(bins[histTIdxs[-1]], hist[histTIdxs[-1]], 'rx')
        plt.title('Histogram of liver density and its class1 = maximal peak (red dot) +-5% of its density (red line).')
        plt.show()

    return class1


def intensity_probability(data, std=20, roi=None, dens_min=10, dens_max=255):
    if roi == None:
        # roi = np.logical_and(data >= dens_min, data <= dens_max)
        roi = np.ones(data.shape, dtype=np.bool)
    voxels = data[np.nonzero(roi)]
    hist, bins = skiexp.histogram(voxels)

    #zeroing histogram outside interval <dens_min, dens_max>
    hist[:dens_min] = 0
    hist[dens_max:] = 0

    max_id = hist.argmax()
    mu = round(bins[max_id])

    prb = scista.norm(loc=mu, scale=std)

    print('liver pdf: mu = %i, std = %i'%(mu, std))

    # plt.figure()
    # plt.plot(bins, hist)
    # plt.hold(True)
    # plt.plot(mu, hist[max_id], 'ro')
    # plt.show()

    probs_L = prb.pdf(voxels)
    probs = np.zeros(data.shape)

    coords = np.argwhere(roi)
    n_elems = coords.shape[0]
    for i in range(n_elems):
        if data.ndim == 3:
            probs[coords[i,0], coords[i,1], coords[i,2]] = probs_L[i]
        else:
            probs[coords[i,0], coords[i,1]] = probs_L[i]

    return probs


def get_zunics_compatness(obj):
    m000 = obj.sum()
    m200 = get_central_moment(obj, 2, 0, 0)
    m020 = get_central_moment(obj, 0, 2, 0)
    m002 = get_central_moment(obj, 0, 0, 2)
    term1 = (3**(5./3)) / (5 * (4*np.pi)**(2./3))
    term2 = m000**(5./3) / (m200 + m020 + m002)
    K = term1 * term2
    return K


def get_central_moment(obj, p, q, r):
    elems = np.argwhere(obj)
    m000 = obj.sum()
    m100 = (elems[:,0]).sum()
    m010 = (elems[:,1]).sum()
    m001 = (elems[:,2]).sum()
    xc = m100 / m000
    yc = m010 / m000
    zc = m001 / m000

    mom = 0
    for el in elems:
        mom += (el[0] - xc)**p + (el[1] - yc)**q + (el[2] - zc)**r

    return mom


def opening3D(data, selem=skimor.disk(3), sliceId=0):
    if sliceId == 0:
        for i in range(data.shape[0]):
            data[i,:,:] = skimor.binary_opening(data[i,:,:], selem)
    elif sliceId == 2:
        for i in range(data.shape[2]):
            data[:,:,i] = skimor.binary_opening(data[:,:,i], selem)
    return data


def closing3D(data, selem=skimor.disk(3), slicewise=False, sliceId=0):
    if slicewise:
        if sliceId == 0:
            for i in range(data.shape[0]):
                data[i, :, :] = skimor.binary_closing(data[i, :, :], selem)
        elif sliceId == 2:
            for i in range(data.shape[2]):
                data[:, :, i] = skimor.binary_closing(data[:, :, i], selem)
    else:
        data = scindimor.binary_closing(data, selem)
    return data


def eroding3D(data, selem=skimor.disk(3), slicewise=False, sliceId=0):
    if slicewise:
        if sliceId == 0:
            for i in range(data.shape[0]):
                data[i, :, :] = skimor.binary_erosion(data[i, :, :], selem)
        elif sliceId == 2:
            for i in range(data.shape[2]):
                data[:, :, i] = skimor.binary_erosion(data[:, :, i], selem)
    else:
        data = scindimor.binary_erosion(data, selem)
    return data


def resize3D(data, scale, sliceId=2, method='cv2'):
    if sliceId == 2:
        n_slices = data.shape[2]
        # new_shape = cv2.resize(data[:,:,0], None, fx=scale, fy=scale).shape
        new_shape = scindiint.zoom(data[:,:,0], scale).shape
        new_data = np.zeros(np.hstack((new_shape,n_slices)), dtype=np.int)
        for i in range(n_slices):
            # new_data[:,:,i] = cv2.resize(data[:,:,i], None, fx=scale, fy=scale)
            # new_data[:,:,i] = (255 * skitra.rescale(data[:,:,0], scale)).astype(np.int)
            if method == 'cv2':
                new_data[:,:,i] = cv2.resize(data[:,:,i], (0,0),  fx=scale, fy=scale, interpolation=cv2.INTER_NEAREST)
            else:
                new_data[:,:,i] = scindiint.zoom(data[:,:,i], scale)
    elif sliceId == 0:
        n_slices = data.shape[0]
        # new_shape = cv2.resize(data[0,:,:], None, fx=scale, fy=scale).shape
        # new_shape = skitra.rescale(data[0,:,:], scale).shape
        new_shape =  scindiint.zoom(data[0,:,:], scale).shape
        new_data = np.zeros(np.hstack((n_slices, new_shape)), dtype=np.int)
        for i in range(n_slices):
            # new_data[i,:,:] = cv2.resize(data[i,:,:], None, fx=scale, fy=scale)
            # new_data[i,:,:] = (255 * skitra.rescale(data[i,:,:], scale)).astype(np.int)
            if method == 'cv2':
                new_data[i,:,:] = cv2.resize(data[i,:,:], (0,0),  fx=scale, fy=scale, interpolation=cv2.INTER_NEAREST)
            else:
                new_data[i,:,:] = scindiint.zoom(data[i,:,:], scale)
    return new_data


def get_overlay(mask, alpha=0.3, color='r'):
    layer = None
    if color == 'r':
        layer = np.dstack((255*mask, np.zeros_like(mask), np.zeros_like(mask), alpha * mask))
    elif color == 'g':
        layer = alpha * np.dstack((np.zeros_like(mask), mask, np.zeros_like(mask)))
    elif color == 'b':
        layer = alpha * np.dstack((np.zeros_like(mask), np.zeros_like(mask), mask))
    elif color == 'c':
        layer = alpha * np.dstack((np.zeros_like(mask), mask, mask))
    elif color == 'm':
        layer = alpha * np.dstack((mask, np.zeros_like(mask), mask))
    elif color == 'y':
        layer = alpha * np.dstack((mask, mask, np.zeros_like(mask)))
    else:
        print 'Unknown color, using red as default.'
        layer = alpha * np.dstack((mask, np.zeros_like(mask), np.zeros_like(mask)))
    return layer


def slim_seeds(seeds, sliceId=2):
    slims = np.zeros_like(seeds)
    if sliceId == 0:
        for i in range(seeds.shape[0]):
            layer = seeds[i,:,:]
            labels = skimor.label(layer, neighbors=4, background=0) + 1
            n_labels = labels.max()
            for o in range(1,n_labels+1):
                centroid = np.round(skimea.regionprops(labels == o)[0].centroid)
                slims[i, centroid[0], centroid[1]] = 1
    return slims


def crop_to_bbox(im, mask):
    if im.ndim == 2:
        # obj_rp = skimea.regionprops(mask.astype(np.integer), properties=('BoundingBox'))
        obj_rp = skimea.regionprops(mask.astype(np.integer))
        bbox = obj_rp[0].bbox # minr, minc, maxr, maxc

        bbox = np.array(bbox)
        # okrajove podminky
        bbox[0] = max(0, bbox[0]-1)
        bbox[1] = max(0, bbox[1]-1)
        bbox[2] = min(im.shape[0], bbox[2]+1)
        bbox[3] = min(im.shape[1], bbox[3]+1)

        # im = im[bbox[0]-1:bbox[2] + 1, bbox[1]-1:bbox[3] + 1]
        # mask = mask[bbox[0]-1:bbox[2] + 1, bbox[1]-1:bbox[3] + 1]
        im = im[bbox[0]:bbox[2], bbox[1]:bbox[3]]
        mask = mask[bbox[0]:bbox[2], bbox[1]:bbox[3]]
    elif im.ndim == 3:
        coords = np.nonzero(mask)
        s_min = max(0, min(coords[0]) - 1)
        s_max = min(im.shape[0], max(coords[0]) + 2)
        r_min = max(0, min(coords[1]) - 1)
        r_max = min(im.shape[1], max(coords[1]) + 2)
        c_min = max(0, min(coords[2]) - 1)
        c_max = min(im.shape[2], max(coords[2]) + 2)

        # im = im[r_min-1:r_max+1, c_min-1:c_max+1, s_min-1:s_max+1]
        # mask = mask[r_min-1:r_max+1, c_min-1:c_max+1, s_min-1:s_min+1]
        im = im[s_min:s_max, r_min:r_max, c_min:c_max]
        mask = mask[s_min:s_max, r_min:r_max, c_min:c_max]

    return im, mask


def slics_3D(im, pseudo_3D=True, n_segments=100, get_slicewise=False):
    if im.ndim != 3:
        raise Exception('3D image is needed.')

    if not pseudo_3D:
        # need to convert to RGB image
        im_rgb = np.zeros((im.shape[0], im.shape[1], im.shape[2], 3))
        im_rgb[:,:,:,0] = im
        im_rgb[:,:,:,1] = im
        im_rgb[:,:,:,2] = im

        suppxls = skiseg.slic(im_rgb, n_segments=n_segments, spacing=(2,1,1))

    else:
        suppxls = np.zeros(im.shape)
        if get_slicewise:
            suppxls_slicewise = np.zeros(im.shape)
        offset = 0
        for i in range(im.shape[0]):
            suppxl = skiseg.slic(cv2.cvtColor(im[i,:,:], cv2.COLOR_GRAY2RGB), n_segments=n_segments)
            suppxls[i,:,:] = suppxl + offset
            if get_slicewise:
                suppxls_slicewise[i,:,:] = suppxl
            offset = suppxls.max() + 1

    if get_slicewise:
        return suppxls, suppxls_slicewise
    else:
        return suppxls


def get_superpxl_intensities(im, suppxls):
    """Calculates mean intensities of pixels in superpixels
    inputs:
        im ... grayscale image, ndarray [MxN]
        suppxls ... image with suppxls labels, ndarray -same shape as im
    outputs:
        suppxl_intens ... vector with suppxls mean intensities
    """
    n_suppxl = np.int(suppxls.max() + 1)
    suppxl_intens = np.zeros(n_suppxl)

    for i in range(n_suppxl):
        sup = suppxls == i
        vals = im[np.nonzero(sup)]
        try:
            suppxl_intens[i] = np.mean(vals)
        except:
            suppxl_intens[i] = -1

    return suppxl_intens


def suppxl_ints2im(suppxls, suppxl_ints=None, im=None):
    """Replaces superpixel labels with their mean value.
    inputs:
        suppxls ... image with suppxls labels, ndarray
        suppxl_intens ... vector with suppxls mean intensities
        im ... input image
    outputs:
        suppxl_ints_im ... image with suppxls mean intensities, ndarray same shape as suppxls
    """

    suppxl_ints_im = np.zeros(suppxls.shape)

    if suppxl_ints is None and im is not None:
        suppxl_ints = get_superpxl_intensities(im, suppxls)

    for i in np.unique(suppxls):
        sup = suppxls == i
        val = suppxl_ints[i]
        suppxl_ints_im[np.nonzero(sup)] = val

    return suppxl_ints_im


def remove_empty_suppxls(suppxls):
    """Remove empty superpixels. Sometimes there are superpixels(labels), which are empty. To overcome subsequent
    problems, these empty superpixels should be removed.
    inputs:
        suppxls ... image with suppxls labels, ndarray [MxN]-same size as im
    outputs:
        new_supps ... image with suppxls labels, ndarray [MxN]-same size as im, empty superpixel labels are removed
    """
    n_suppxls = np.int(suppxls.max() + 1)
    new_supps = np.zeros(suppxls.shape, dtype=np.integer)
    idx = 0
    for i in range(n_suppxls):
        sup = suppxls == i
        if sup.any():
            new_supps[np.nonzero(sup)] = idx
            idx += 1
    return new_supps


def label_3D(data, class_labels, background=-1):
    # class_labels = np.unique(data[data > background])
    labels = - np.ones(data.shape, dtype=np.int)
    curr_l = 0
    for c in class_labels:
        x = data == c
        labs, n_labels = scindimea.label(x)
        print 'labels: ', np.unique(labs)
        # py3DSeedEditor.py3DSeedEditor(labs).show()
        for l in range(n_labels + 1):
            labels = np.where(labs == l, curr_l, labels)
            curr_l += 1
    print 'min = %i, max = %i' % (labels.min(), labels.max())
    return labels


def load_pickle_data(fname, win_level=50, win_width=350, slice_idx=-1):
    fcontent = None
    try:
        import gzip
        f = gzip.open(fname, 'rb')
        fcontent = f.read()
        f.close()
    except Exception as e:
        f = open(fname, 'rb')
        fcontent = f.read()
        f.close()
    data_dict = pickle.loads(fcontent)

    data = windowing(data_dict['data3d'], level=win_level, width=win_width)

    mask = data_dict['segmentation']

    voxel_size = data_dict['voxelsize_mm']


    if slice_idx != -1:
        data = data[slice_idx, :, :]
        mask = mask[slice_idx, :, :]

    return data, mask, voxel_size