| import os |
| import numpy as np |
| from PIL import Image |
|
|
| data_root = './MRNet/MRNet-v1.0' |
| output_root = os.path.join(data_root, '{split}_slides') |
|
|
| splits = ['train', 'valid'] |
| views = ['axial', 'coronal', 'sagittal'] |
| num_slices_to_keep = 10 |
|
|
| for split in splits: |
| for view in views: |
| input_dir = os.path.join(data_root, split, view) |
| output_dir = os.path.join(data_root, f'{split}_slides', view) |
|
|
| if not os.path.isdir(input_dir): |
| print(f"Directory not found, skipping: {input_dir}") |
| continue |
|
|
| os.makedirs(output_dir, exist_ok=True) |
| npy_files = sorted([f for f in os.listdir(input_dir) if f.endswith('.npy')]) |
| print(f"[{split}/{view}] Found {len(npy_files)} .npy files, processing...") |
|
|
| for filename in npy_files: |
| file_path = os.path.join(input_dir, filename) |
| volume_data = np.load(file_path) |
|
|
| sample_name = os.path.splitext(filename)[0] |
| sample_output_dir = os.path.join(output_dir, sample_name) |
| os.makedirs(sample_output_dir, exist_ok=True) |
|
|
| total_slices = volume_data.shape[0] |
|
|
| if total_slices <= num_slices_to_keep: |
| selected_indices = np.arange(total_slices) |
| else: |
| selected_indices = np.linspace(0, total_slices - 1, num_slices_to_keep, dtype=int) |
|
|
| for save_idx, slice_idx in enumerate(selected_indices): |
| slice_2d = volume_data[slice_idx, :, :] |
|
|
| if slice_2d.dtype != np.uint8: |
| slice_min = slice_2d.min() |
| slice_max = slice_2d.max() |
| if slice_max > slice_min: |
| slice_2d = (slice_2d - slice_min) / (slice_max - slice_min) * 255.0 |
| else: |
| slice_2d = np.zeros_like(slice_2d) |
| slice_2d = slice_2d.astype(np.uint8) |
|
|
| img = Image.fromarray(slice_2d) |
| save_path = os.path.join( |
| sample_output_dir, |
| f'slice_{save_idx:02d}_orig_{slice_idx:03d}.png' |
| ) |
| img.save(save_path) |
|
|
| print(f" Converted: {sample_name} (original {total_slices} slices, saved {len(selected_indices)})") |
|
|
| print("\nAll done!") |
|
|
