纹理融合框架分类儿童肺炎胸片准确率0.9818：《科学报告》用5856张影像、SE融合F1值0.9875，11月13日刊出
《科学报告》2026年11月13日刊出一项研究，评估一个双分支深度与纹理融合框架在儿童肺炎胸部X光影像分类上的表现。框架中，EfficientNet-B0负责编码原始影像，ResNet-18负责编码三张灰度共生矩阵图，即对比度、相异性与熵，再由挤压激励注意力共同校准拼接后的特征后作分类。作者为Kantasa等人，论文数字对象标识为10.1038/s41598-026-78453-x，2026年3月17日收稿、10月26日接受、11月13日刊出。
研究把5856张影像的数据集重新划分为4099张训练、877张验证与880张测试影像，所有配置使用相同的划分。期刊摘要报告，挤压激励融合取得的准确率为0.9818，F1值为0.9875，曲线下面积为0.9958，马修斯相关系数为0.9539，其准确率比单独的原始EfficientNet-B0多对5个预测、比无门控融合多对6个。
作者同时明确了这项工作的限度。纹理门的平均归一化值为0.0759，但作者说明这一数值不是因果重要性的度量，其相对更简单辅助特征的收益尚未检验。归因与扰动分析属模型审计，没有放射科医生验证的定位结果。研究只用了一次影像级划分，没有外部验证或重复交叉验证，患者的独立性无法核实。作者据此表示，结果支持在这个基准数据集上继续研究，但不确立临床泛化能力或部署适用性。
（来源：《科学报告》）
A Texture Fusion Framework Classifies Pediatric Pneumonia Chest Radiographs at 0.9818 Accuracy: 5,856 Images and an SE Fusion F1 Score of 0.9875 in Scientific Reports, Published November 13
A study published in Scientific Reports on November 13, 2026, evaluated a two-branch deep and texture fusion framework for pediatric pneumonia classification from chest radiographs. EfficientNet-B0 encodes the raw image, while ResNet-18 encodes three gray-level co-occurrence matrix maps, namely contrast, dissimilarity, and entropy, and squeeze-and-excitation attention jointly recalibrates the concatenated embeddings before classification. The authors are Kantasa and colleagues, and the paper carries digital object identifier 10.1038/s41598-026-78453-x, was received on March 17, 2026, accepted on October 26, 2026, and published on November 13, 2026. The 5,856-image dataset was repartitioned into 4,099 training, 877 validation, and 880 test images, using identical assignments for all configurations. The journal abstract reported that squeeze-and-excitation fusion achieved accuracy of 0.9818, an F1-score of 0.9875, an area under the receiver operating curve of 0.9958, and a Matthews correlation coefficient of 0.9539, exceeding raw EfficientNet-B0 by five correct predictions and ungated fusion by six. The authors stated the limits plainly. The mean normalized texture gate was 0.0759, but that summary is not a measure of causal importance, and the benefit over simpler auxiliary features remains untested. Attribution and perturbation analyses are model audits, without radiologist-validated localization. The study uses one image-level split, with no external or repeated cross-validation, and patient independence cannot be verified. The authors concluded that the results support further investigation on this benchmark but do not establish clinical generalizability or deployment suitability.
(Source: Scientific)
来源：《科学报告》
发布：GSM·美国（发布链接）
图片来源：图：一张显示肺部肺炎病灶的胸部X光影像。Mikael Häggström，CC0，经 Wikimedia Commons。 / Photo: Mikael Häggström, CC0, via Wikimedia Commons.
GSM Global · 本篇英文版 English edition of this story
