利用 python/keras/tensorflow 计算 DICE系数，评估 3D 分割结果

最新推荐文章于 2024-03-22 16:26:00 发布

原创最新推荐文章于 2024-03-22 16:26:00 发布 · 3.9k 阅读

17 ·

CC 4.0 BY-SA版权

文章标签：

#DICE #python #segmentation

Python应用专栏收录该内容

15 篇文章

订阅专栏

本文介绍了一种使用Python和Keras计算200幅分割图像与对应groundtruth之间的DICE系数的方法，通过评估分割结果的准确性，为图像处理和计算机视觉任务提供量化指标。

有 200幅分割图像和其对应的 ground truth，现利用 python 语言计算 DICE 系数，评估分割结果

import nibabel as nib
import scipy.io as io
import os
import numpy as np
import tensorflow as tf
from keras import backend as K


def dice_coefficient(y_true, y_pred, smooth=0.00001):
    y_true_f = K.flatten(y_true)
    y_pred_f = K.flatten(y_pred)
    intersection = K.sum(y_true_f * y_pred_f)
    return (2. * intersection + smooth) / (K.sum(y_true_f) + K.sum(y_pred_f) + smooth)

pred_dir = '/home/share/name/20200206_test_labeled/code/pred_result'
true_dir = '/home/share/name/20200206_test_labeled/mask'
pred_filenames = os.listdir(pred_dir)
pred_filenames.sort(key=lambda x:x[:-9])

true_filenames = os.listdir(true_dir)
true_filenames.sort(key=lambda x:x[:-8])

dice_value = np.zeros(200)
temp = []
for f in range(200):
    pred_path = os.path.join(pred_dir, pred_filenames[f])
    img_pred = nib.load(pred_path)
    y_pred = img_pred.get_fdata()
    true_path = os.path.join(true_dir, true_filenames[f])
    img_true = nib.load(true_path)
    y_true = img_true.get_fdata()
    temp.append(dice_coefficient(y_true, y_pred))
    pass


with tf.Session() as sess:
    dice_value = sess.run(temp)
    pass

mat_path = '/home/share/name/20200206_test_labeled/DICE_value.mat'
io.savemat(mat_path, {'DICE_value': dice_value})