Code/divide_dataset.py

import os
import json
import shutil
from pathlib import Path

def convert_to_nnunet_format(
    input_images_path, 
    input_labels_path, 
    output_path, 
    dataset_name="Dataset001_pelvic",
    train_ratio=0.8,
    test_ratio=0.2
):
    """
    将CT文件和标签转换为nnU-Net格式
    
    Args:
        input_images_path: CT文件路径 (路径A)
        input_labels_path: 标签文件路径 (路径B)
        output_path: 输出路径
        dataset_name: 数据集名称
        train_ratio: 训练集比例
        test_ratio: 测试集比例
    """
    
    # 创建输出目录结构
    dataset_path = Path(output_path) / dataset_name
    imagesTr_path = dataset_path / "imagesTr"
    labelsTr_path = dataset_path / "labelsTr"
    imagesTs_path = dataset_path / "imagesTs"
    
    # 创建目录
    imagesTr_path.mkdir(parents=True, exist_ok=True)
    labelsTr_path.mkdir(parents=True, exist_ok=True)
    imagesTs_path.mkdir(parents=True, exist_ok=True)
    
    # 获取所有.mha文件
    input_images = Path(input_images_path)
    input_labels = Path(input_labels_path)
    
    image_files = sorted(list(input_images.glob("*.mha")))
    print(f"找到 {len(image_files)} 个图像文件")
    
    if len(image_files) == 0:
        raise ValueError(f"在路径 {input_images_path} 中没有找到.mha文件")
    
    # 验证对应的标签文件是否存在
    valid_pairs = []
    for img_file in image_files:
        # 假设标签文件与图像文件同名
        label_file = input_labels / img_file.name
        if label_file.exists():
            valid_pairs.append((img_file, label_file))
        else:
            print(f"警告: 图像文件 {img_file.name} 没有对应的标签文件")
    
    print(f"找到 {len(valid_pairs)} 对有效的图像-标签文件")
    
    if len(valid_pairs) == 0:
        raise ValueError("没有找到有效的图像-标签文件对")
    
    # 计算训练集和测试集数量
    total_files = len(valid_pairs)
    num_train = int(total_files * train_ratio)
    num_test = total_files - num_train
    
    print(f"训练集: {num_train} 个文件")
    print(f"测试集: {num_test} 个文件")
    
    # 复制文件并重命名
    train_cases = []
    test_cases = []
    
    for i, (img_file, label_file) in enumerate(valid_pairs):
        # nnU-Net命名格式: case_identifier_0000.nii.gz
        case_id = f"{dataset_name.split('_')[1]}_{i:03d}"
        
        if i < num_train:
            # 训练集
            # 复制图像文件
            dst_img = imagesTr_path / f"{case_id}_0000.mha"
            shutil.copy2(img_file, dst_img)
            
            # 复制标签文件
            dst_label = labelsTr_path / f"{case_id}.mha"
            shutil.copy2(label_file, dst_label)
            
            train_cases.append(case_id)
            
        else:
            # 测试集
            dst_img = imagesTs_path / f"{case_id}_0000.mha"
            shutil.copy2(img_file, dst_img)
            
            test_cases.append(case_id)
    
    print(f"文件复制完成")
    print(f"训练集文件: {len(train_cases)} 个")
    print(f"测试集文件: {len(test_cases)} 个")
    
    # 创建dataset.json
    dataset_json = {
        "channel_names": {
            "0": "CT"
        },
        # "labels": {
        #     "background": 0,
        #     "middle": 1,
        #     "right": 2,
        #     "left": 3
        # },
        "labels": {
            "background": 0,
            "1": 1,
            "2": 2,
            "3": 3,
            "4": 4,
            "5": 5,
            "6": 6,
            "7": 7,
            "8": 8,
            "9": 9,
            "10": 10,
            "11": 11
        },
        "numTraining": len(train_cases),
        "file_ending": ".mha",
        "dataset_name": dataset_name,
        "reference": "Pelvic CT Dataset",
        "description": "Pelvic CT segmentation dataset",
        "tensorImageSize": "4D",
        "modality": {
            "0": "CT"
        },
        "dim": 3
    }
    
    # 如果你有多个标签类别，请修改labels字段，例如：
    # "labels": {
    #     "background": 0,
    #     "bone": 1,
    #     "muscle": 2,
    #     "organ": 3
    # }
    
    # 保存dataset.json
    json_path = dataset_path / "dataset.json"
    with open(json_path, 'w', encoding='utf-8') as f:
        json.dump(dataset_json, f, indent=4, ensure_ascii=False)
    
    print(f"dataset.json 已创建: {json_path}")
    
    # 打印最终目录结构
    print("\n创建的目录结构:")
    print(f"{dataset_path}/")
    print(f"├── dataset.json")
    print(f"├── imagesTr/ ({len(train_cases)} files)")
    print(f"├── imagesTs/ ({len(test_cases)} files)")
    print(f"└── labelsTr/ ({len(train_cases)} files)")
    
    return dataset_path

def main():
    # 直接定义路径参数
    # images_path = "/research/phd_y3/pelvic_project/Data/images"
    # labels_path = "/research/phd_y3/pelvic_project/Data/3_part_labels"
    # output_path = "/research/phd_y3/pelvic_project/Code/for_nnUNet/nnUNet_raw_data"
    # dataset_name = "Dataset001_pelvic_three_parts"
    
    # images_path = "/research/phd_y3/pelvic_project/Data/mid_sacrum_bbox_images"
    # labels_path = "/research/phd_y3/pelvic_project/Data/mid_sacrum_bbox_labels"
    # output_path = "/research/phd_y3/pelvic_project/Code/for_nnUNet/nnUNet_raw_data"
    # dataset_name = "Dataset002_mid_sacrum"
    
    # images_path = "/research/phd_y3/pelvic_project/Data/right_hip_bbox_images"
    # labels_path = "/research/phd_y3/pelvic_project/Data/right_hip_bbox_labels"
    # output_path = "/research/phd_y3/pelvic_project/Code/for_nnUNet/nnUNet_raw_data"
    # dataset_name = "Dataset003_right_hip"
    
    # images_path = "/research/phd_y3/pelvic_project/Data/left_hip_bbox_images"
    # labels_path = "/research/phd_y3/pelvic_project/Data/left_hip_bbox_labels"
    # output_path = "/research/phd_y3/pelvic_project/Code/for_nnUNet/nnUNet_raw_data"
    # dataset_name = "Dataset004_left_hip"
    
    images_path = "/research/phd_y3/pelvic_project/Data/images"
    labels_path = "/research/phd_y3/pelvic_project/Data/organ+3_part_label_fixed"
    output_path = "/research/phd_y3/pelvic_project/Code/for_nnUNet/nnUNet_raw_data"
    dataset_name = "Dataset005_organ_pelvic"
    
    train_ratio = 0.7
    test_ratio = 0.3
    
    print(f"开始处理数据集转换:")
    print(f"图像路径: {images_path}")
    print(f"标签路径: {labels_path}")
    print(f"输出路径: {output_path}")
    print(f"数据集名称: {dataset_name}")
    print(f"训练集比例: {train_ratio}")
    print(f"测试集比例: {test_ratio}")
    print("-" * 50)
    
    # 验证输入路径
    if not os.path.exists(images_path):
        print(f"❌ 图像路径不存在: {images_path}")
        return 1
    
    if not os.path.exists(labels_path):
        print(f"❌ 标签路径不存在: {labels_path}")
        return 1
    
    # 转换数据
    try:
        result_path = convert_to_nnunet_format(
            images_path, 
            labels_path, 
            output_path, 
            dataset_name,
            train_ratio,
            test_ratio
        )
        print(f"\n✅ 转换完成! 数据集保存在: {result_path}")
        print(f"\n下一步:")
        print(f"1. 设置环境变量: export nnUNet_raw_data_base='{Path(output_path).absolute()}'")
        print(f"2. 运行nnU-Net预处理: nnUNet_plan_and_preprocess -t {dataset_name.split('_')[0].replace('Dataset', '')}")
        
    except Exception as e:
        print(f"❌ 转换失败: {str(e)}")
        return 1
    
    return 0

if __name__ == "__main__":
    exit(main())