Replication Data and Code for "Phase-Domain Radiofrequency Signatures for Non-Invasive Bone Fracture Detection: A Two-Dimensional Computational Proof-of-Concept Study Using FDTD Simulation and Machine Learning

This repository contains all materials required to reproduce the results in the paper "Phase-Domain Radiofrequency Signatures for Non-Invasive Bone Fracture Detection: A Two-Dimensional Computational Proof-of-Concept Study Using FDTD Simulation and Machine Learning" (v20). It includes: Pre-computed 486-dimensional feature matrices for the original 27,088 samples Monte Carlo healthy class expansion (50 anatomically varied configurations, +1,600 samples) Group-wise 5-fold cross-validation fold assignments (no leakage) Monte Carlo generation and re-evaluation scripts Key results files (Before/After Monte Carlo comparison) All figures and supplementary tables The dataset enables independent reproduction of all classification results (AUC = 0.943 after MC expansion, specificity = 86.2%).本仓库包含复现论文《面向无创骨折检测的相域射频特征：基于时域有限差分（Finite-Difference Time-Domain, FDTD）仿真与机器学习的二维计算概念验证研究》（v20版）全部结果所需的所有材料。 其包含如下内容： - 原始27088个样本的预计算486维特征矩阵 - 蒙特卡洛（Monte Carlo）健康类别扩展（50种解剖结构多样化配置，新增1600个样本） - 分组式5折交叉验证折次分配方案（无数据泄露） - 蒙特卡洛生成与重评估脚本 - 核心结果文件（蒙特卡洛扩展前后对比结果） - 全部图表与补充表格 本数据集支持独立复现全部分类结果（蒙特卡洛扩展后AUC（Area Under the Receiver Operating Characteristic Curve）为0.943，特异度为86.2%）。