braST_train.py

import os
import nibabel as nib
import numpy as np
import matplotlib.pyplot as plt
import pandas as pd

# 输入路径和输出路径
input_folder = "training_data1_v2"
output_folder = "BraTS"

# 定义模态后缀和目标文件夹的模态名称
modalities = {"T1": "t1n", "T1c": "t1c", "T2": "t2w", "FLAIR": "t2f"}

# 创建输出文件夹结构
os.makedirs(os.path.join(output_folder, "train", "image"), exist_ok=True)


# 数据归一化函数（修正除零问题）
def normalize_slice(slice_data):
    if np.max(slice_data) == 0:
        return np.zeros(slice_data.shape, dtype=np.uint8)  # 全零切片
    slice_data = slice_data - np.min(slice_data)  # 移动到 0
    slice_data = slice_data / np.max(slice_data)  # 归一化到 [0, 1]
    slice_data = (slice_data * 255).astype(np.uint8)  # 转换到 [0, 255]
    return slice_data


# 标签分配函数
def assign_label(patient_folder):
    seg_file = os.path.join(patient_folder, f"{os.path.basename(patient_folder)}-seg.nii.gz")
    if not os.path.exists(seg_file):
        print(f"分割文件缺失: {seg_file}")
        return None  # 如果分割文件缺失，返回 None

    seg_img = nib.load(seg_file)
    seg_data = seg_img.get_fdata()
    unique_labels = np.unique(seg_data)

    if len(unique_labels) == 1 and unique_labels[0] == 0:
        return 0  # 健康人
    elif 1 in unique_labels:
        return 2  # 存在坏死区域
    else:
        return 1  # 水肿和增强组织为主


# 处理患者文件夹
def process_patient(patient_folder, patient_id, label):
    patient_output_folder = os.path.join(output_folder, "train", "image", f"{patient_id:03d}_{label}")
    os.makedirs(patient_output_folder, exist_ok=True)

    # 遍历模态
    for modality, suffix in modalities.items():
        modality_folder = os.path.join(patient_output_folder, modality)
        os.makedirs(modality_folder, exist_ok=True)

        # 构造文件路径
        nii_file = os.path.join(patient_folder, f"{os.path.basename(patient_folder)}-{suffix}.nii.gz")
        if not os.path.exists(nii_file):
            print(f"缺失文件: {nii_file}")
            continue

        # 加载 NIfTI 文件
        img = nib.load(nii_file)
        data = img.get_fdata()

        # 遍历切片
        for i in range(data.shape[2]):  # 假设第三维是切片维度
            slice_data = data[:, :, i]  # 提取切片
            slice_data = normalize_slice(slice_data)  # 归一化

            # 保存为 PNG
            slice_path = os.path.join(modality_folder, f"slice_{i:03d}.png")
            plt.imsave(slice_path, slice_data, cmap="gray")
    print(f"患者 {patient_id} 数据处理完成！")


# 遍历患者文件夹并处理
patient_folders = [os.path.join(input_folder, folder) for folder in os.listdir(input_folder) if
                   os.path.isdir(os.path.join(input_folder, folder))]
csv_data = []

for patient_id, patient_folder in enumerate(patient_folders, start=1):
    label = assign_label(patient_folder)  # 分配标签
    if label is None:
        continue  # 跳过没有分割文件的患者

    process_patient(patient_folder, patient_id, label)

    # 记录到 CSV
    csv_data.append({"ID": patient_id, "Label": label})

# 保存标签 CSV 文件
csv_path = os.path.join(output_folder, "train", "train_labels.csv")
pd.DataFrame(csv_data).to_csv(csv_path, index=False)
print(f"标签文件已保存到: {csv_path}")
print("所有患者数据处理完成！")