Lung Cancer Images for Classification

The author obtained the CT Scan images dataset from the Cancer Imaging Archive collection; there are two related research on the dataset: "LDCT and Projection Image" (https://doi.org/10.7937/9NPB-2637) and the "Quantitative computed tomographic descriptors associate tumor shape complexity and intratumor heterogeneity with prognosis in lung adenocarcinoma" (https://doi.org/10.7937/K9/TCIA.2015.A6V7JIWX). The research combines the series of filters named "Multi-stage" and deep learning CNN for classification. The 3D lung images would be extracted and filtered out using geometric properties from the dataset on a smaller dimension of 20x20x20 before the training. The intended output is the cancer annotation (as the CNN output) and RADS scores (from the nodule properties). 本研究的研究者从癌症影像档案库（Cancer Imaging Archive）中获取了本次所用的CT扫描影像数据集；针对该数据集已有两项相关研究：其一为《LDCT与投影影像》（https://doi.org/10.7937/9NPB-2637），其二为《定量计算机断层扫描特征将肺腺癌的肿瘤形态复杂度、瘤内异质性与患者预后相关联》（https://doi.org/10.7937/K9/TCIA.2015.A6V7JIWX）。本研究结合名为“多阶段”的系列滤波器与深度学习卷积神经网络（Convolutional Neural Network）开展分类任务。训练前，将从数据集中提取三维肺部影像，并通过几何属性筛选，将其维度压缩至20×20×20的小尺寸数据。本次任务的预期输出为癌症标注（作为卷积神经网络的输出结果）以及基于结节属性得到的RADS评分。

搜集汇总

背景与挑战

背景概述

该数据集包含来自Cancer Imaging Archive的CT扫描图像，用于肺癌分类研究，结合多阶段过滤器和深度学习CNN方法，预处理中从20x20x20维度提取3D肺部图像，输出包括癌症注释和RADS评分。