基于深度学习的腹内脂肪分割算法DiMax

利用Active contour 算法识别腹部脂肪像素点；对识别出的腹部脂肪像素点采用多尺度块作为特征输入，通过新的深度神经网络算法自动学习分层的抽象本质特征，将这些特征输入分类算法得到初步分割结果；然后将初步分割结果转到极坐标中，利用 SAT 在极坐标下成为图像底部的特征，得到腹部脂肪分割图；最后按照梯度高低对腹部脂肪分割图中各类型脂肪比例进行计算，并通过体绘制技术直观显示计算结果。

Identify abdominal fat pixels using the Active Contour algorithm. Take the identified abdominal fat pixels with multi-scale patches as feature inputs, and automatically learn hierarchical abstract essential features via a novel deep neural network algorithm. Input these features into a classification algorithm to obtain preliminary segmentation results. Then transform the preliminary segmentation results into polar coordinates, and utilize the characteristic that SAT appears at the bottom of the image in polar coordinates to generate the abdominal fat segmentation map. Finally, calculate the proportion of each type of fat in the abdominal fat segmentation map according to the gradient magnitude, and visually display the calculation results via volume rendering technology.

搜集汇总

背景与挑战

背景概述

该数据集名为'基于深度学习的腹内脂肪分割算法DiMax'，主要用于医疗健康领域，通过深度学习和Active contour算法实现腹内脂肪的自动分割和计算。数据集大小为2GB，更新频率为每年一次，覆盖全国范围。