Trained FIL model

The FIL model was trained on all 15 training subjects (fil15) from the MICCAI Challenge dataset, using a minimally informative, but proper Wishart prior, with <i>v</i>₀=1.0. An augmentation search radius of 3 voxels was used with a Gaussian weighting standard deviation of 2.0 voxels. Patch sizes were 4x4x4 voxels, and four outer iterations were used for model training. Up to <i>K</i>=24 basis functions were available to encode each patch. This file is the result.<br>A second model was trained on all 30 subjects (fil30) from the MICCAI Challenge dataset (35 scans, with repeat scans given a weighting of 0.5). Augmentation search radius was 2 voxels, with a Gaussian weighting standard deviation of 1.5 voxels. Other parameters remained the same as for fil15.<br>

本FIL模型基于MICCAI挑战赛数据集（MICCAI Challenge Dataset）中的全部15个训练受试者（fil15）完成训练，采用了最小信息且恰当的威沙特（Wishart）先验，其自由度参数v₀=1.0。训练设置了3个体素的增强搜索半径，高斯加权标准差为2.0个体素；训练块尺寸为4×4×4体素，模型训练共执行4轮外层迭代。每个图像块可使用最多K=24个基函数进行编码。本文件即为该模型的训练结果。
第二个模型同样基于MICCAI挑战赛数据集的全部30个受试者（fil30）完成训练，该数据集共包含35次扫描，其中重复扫描的权重设为0.5。本次训练的增强搜索半径为2个体素，高斯加权标准差为1.5个体素，其余参数与fil15模型保持一致。