CT/lung/src/preprocessing/build_crops.py

from __future__ import annotations

import argparse
from pathlib import Path

import numpy as np

from src.datasets.common import load_array


def center_crop_or_pad(volume: np.ndarray, center: tuple[int, int, int], size: tuple[int, int, int]) -> np.ndarray:
    output = np.zeros(size, dtype=volume.dtype)
    src_slices = []
    dst_slices = []
    for axis, crop_size in enumerate(size):
        start = int(round(center[axis] - crop_size / 2))
        end = start + crop_size
        src_start = max(start, 0)
        src_end = min(end, volume.shape[axis])
        dst_start = max(-start, 0)
        dst_end = dst_start + (src_end - src_start)
        src_slices.append(slice(src_start, src_end))
        dst_slices.append(slice(dst_start, dst_end))
    output[tuple(dst_slices)] = volume[tuple(src_slices)]
    return output


def crop_from_mask(volume: np.ndarray, mask: np.ndarray, size: tuple[int, int, int]) -> tuple[np.ndarray, tuple[int, int, int]]:
    coords = np.argwhere(mask > 0)
    if len(coords) == 0:
        center = tuple(int(s // 2) for s in volume.shape)
    else:
        center = tuple(int(v) for v in coords.mean(axis=0))
    return center_crop_or_pad(volume, center, size), center


def main() -> None:
    parser = argparse.ArgumentParser(description="Build a single lesion-centered NPZ crop.")
    parser.add_argument("--ct", required=True)
    parser.add_argument("--mask", required=True)
    parser.add_argument("--out", required=True)
    parser.add_argument("--size", nargs=3, type=int, default=(96, 96, 96))
    args = parser.parse_args()

    ct = load_array(args.ct, key="ct")
    mask = load_array(args.mask, key="mask")
    crop, center = crop_from_mask(ct, mask, tuple(args.size))
    mask_crop = center_crop_or_pad(mask, center, tuple(args.size))
    Path(args.out).parent.mkdir(parents=True, exist_ok=True)
    np.savez_compressed(args.out, ct=crop.astype(np.float32), mask=mask_crop.astype(np.float32), center=np.array(center))


if __name__ == "__main__":
    main()