Spaces:
Sleeping
Sleeping
File size: 14,092 Bytes
377b913 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 | #!/usr/bin/env python
# -*- coding: utf-8 -*-
# File: prepare_data.py
#
# Turn an arbitrary tree of DICOM data into the folder layout the RL-Medical /
# PINN pipeline expects:
#
# data/
# <anything>/.../<dicom slices> <-- your raw data, ANY structure
# images/ <-- generated: one <name>.nii.gz per DICOM series
# filenames/ <-- generated: image_files.txt, landmark_files.txt
# landmarks/ <-- generated: one <name>.txt per volume (>=15 points)
# models/ <-- (left untouched)
#
# It walks everything under --data recursively, groups DICOM files into series
# (3D volumes), converts each to .nii.gz, and writes the filenames + landmark
# files so `dataReader.py` / `medical.py` run unchanged.
#
# Backends (auto-detected, best first):
# 1. SimpleITK -- robust series grouping + ordering (recommended; the repo
# already depends on it).
# 2. pydicom (+ mini_nifti writer)
# 3. mini_dicom (built-in, dependency-free fallback)
#
# LANDMARKS: you almost certainly do NOT have annotations yet. This script
# writes VALID PLACEHOLDER landmark files (>=15 points, correct format) so the
# pipeline is runnable immediately, and prints a loud warning. Replace them with
# real annotations before training a detector, or pass --landmarks-csv to fill
# in known points. (For the physics/PINN visualisations, the placeholder seed is
# enough to compute a geodesic field.)
#
# Usage:
# python prepare_data.py --data /path/to/data
# python prepare_data.py --data ./data --num-landmarks 15 --landmarks centroid
# python prepare_data.py --data ./data --landmarks-csv my_points.csv
#
# CSV format for --landmarks-csv (one row per point):
# volume_name,index,x,y,z
# 2_MR5OOO_20250514_1,13,89,88,84 # AC point for that volume
# 2_MR5OOO_20250514_1,14,88,93,77 # PC point
import os
import sys
import argparse
import numpy as np
sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
# ---- optional backends ------------------------------------------------------
try:
import SimpleITK as sitk
_HAVE_SITK = True
except Exception:
_HAVE_SITK = False
try:
import pydicom
_HAVE_PYDICOM = True
except Exception:
_HAVE_PYDICOM = False
import mini_dicom
from mini_nifti import write_nii
# =============================================================================
# Series discovery -- works on ANY folder structure
# =============================================================================
def find_dicom_dirs(root, images_dirname):
"""Yield directories under `root` that contain at least one DICOM file,
skipping the generated output folders."""
skip = {images_dirname, "filenames", "landmarks", "models"}
for dirpath, dirnames, filenames in os.walk(root):
rel = os.path.relpath(dirpath, root)
parts = set(rel.split(os.sep))
if parts & skip:
continue
dcm = [os.path.join(dirpath, f) for f in filenames
if _looks_dicom(os.path.join(dirpath, f))]
if dcm:
yield dirpath, dcm
def _looks_dicom(path):
if not os.path.isfile(path):
return False
low = path.lower()
if low.endswith((".dcm", ".ima")):
return True
if low.endswith((".nii", ".nii.gz", ".txt", ".json", ".png", ".gif", ".pt")):
return False
return mini_dicom.is_dicom(path) # DICM magic at byte 128 (extension-less ok)
def group_series_sitk(dirpath):
"""Return {series_uid: [ordered file paths]} using GDCM (robust)."""
reader = sitk.ImageSeriesReader()
ids = reader.GetGDCMSeriesIDs(dirpath)
out = {}
if ids:
for sid in ids:
files = reader.GetGDCMSeriesFileNames(dirpath, sid)
if files:
out[sid] = list(files)
return out
def group_series_fallback(dcm_files):
"""Group by SeriesInstanceUID via mini_dicom; if unreadable, one series/dir."""
groups = {}
for p in dcm_files:
try:
d = mini_dicom.read_dicom(p)
sid = d.get("SeriesInstanceUID") or "series"
except Exception:
sid = "series"
groups.setdefault(sid, []).append(p)
return groups
# =============================================================================
# Reading a series into a volume + spacing
# =============================================================================
def load_series(files, dirpath):
if _HAVE_SITK:
reader = sitk.ImageSeriesReader()
reader.SetFileNames(files)
img = reader.Execute()
vol = sitk.GetArrayFromImage(img) # (z, y, x)
vol = np.transpose(vol, (2, 1, 0)).astype(np.float32) # -> (x, y, z)
return vol, tuple(img.GetSpacing()) # (sx, sy, sz)
if _HAVE_PYDICOM:
slices = []
for p in files:
ds = pydicom.dcmread(p)
slices.append(ds)
slices.sort(key=lambda s: float(getattr(s, "InstanceNumber", 0)))
vol = np.stack([s.pixel_array for s in slices], axis=-1) # (rows,cols,z)=(y,x,z)
vol = np.transpose(vol, (1, 0, 2)).astype(np.float32) # -> (x,y,z)
ps = [float(x) for x in getattr(slices[0], "PixelSpacing", [1, 1])]
st = float(getattr(slices[0], "SliceThickness", 1) or 1)
return vol, (ps[1], ps[0], st)
return mini_dicom.volume_from_series(files)
# =============================================================================
# Landmark file generation
# =============================================================================
def make_landmarks(vol, mode, n, provided):
"""Return an (n,3) array of landmark voxel coords.
provided: {index: (x,y,z)} of known points for THIS volume (from CSV).
Unknown indices are filled with a placeholder (center or intensity centroid).
"""
if mode == "centroid":
thr = np.percentile(vol, 60)
mask = vol > thr
if mask.any():
xs, ys, zs = np.where(mask)
base = np.array([xs.mean(), ys.mean(), zs.mean()])
else:
base = np.array(vol.shape) / 2.0
else: # center
base = np.array(vol.shape) / 2.0
pts = np.tile(base, (n, 1))
# scatter placeholders slightly so they're not all identical
offs = np.linspace(-6, 6, n)
pts[:, 0] += offs
for idx, xyz in (provided or {}).items():
if 0 <= idx < n:
pts[idx] = xyz
return np.clip(np.round(pts), 0, np.array(vol.shape) - 1).astype(int)
def _series_cosines(files):
"""Read ImageOrientationPatient (row_dir, col_dir) from the first slice.
Returns None if unavailable (auto_landmarks then assumes axial identity)."""
for p in files:
try:
if _HAVE_PYDICOM:
ds = pydicom.dcmread(p, stop_before_pixels=True)
iop = [float(x) for x in ds.ImageOrientationPatient]
else:
d = mini_dicom.read_dicom(p)
iop = [float(x) for x in d["ImageOrientationPatient"].split("\\")]
return (iop[:3], iop[3:])
except Exception:
continue
return None
def load_csv(path):
"""Parse --landmarks-csv into {volume_name: {index: (x,y,z)}}."""
table = {}
if not path:
return table
with open(path) as f:
for line in f:
line = line.strip()
if not line or line.lower().startswith("volume"):
continue
parts = [p.strip() for p in line.split(",")]
if len(parts) < 5:
continue
name, idx, x, y, z = parts[0], int(parts[1]), float(parts[2]), float(parts[3]), float(parts[4])
table.setdefault(name, {})[idx] = (x, y, z)
return table
# =============================================================================
# Main
# =============================================================================
def sanitize(root, dirpath, series_key, used):
"""Build a clean, unique volume name from the path (structure-agnostic)."""
rel = os.path.relpath(dirpath, root)
parts = [p for p in rel.split(os.sep) if p and p.upper() != "DICOM"]
name = "_".join(parts) if parts else "vol"
# disambiguate multiple series in the same dir
if series_key and series_key != "series":
name += "_s" + series_key.split(".")[-1][-4:]
name = "".join(c if (c.isalnum() or c in "_-") else "_" for c in name)
base = name
k = 1
while name in used:
name = f"{base}_{k}"; k += 1
used.add(name)
return name
def main():
ap = argparse.ArgumentParser(description="Prepare DICOM data for RL-Medical/PINN.")
ap.add_argument("--data", required=True, help="path to the top-level data/ folder")
ap.add_argument("--images-dir", default="images", help="output folder for .nii.gz")
ap.add_argument("--num-landmarks", type=int, default=15,
help=">=15 required (indices 13=AC, 14=PC are used by the code)")
ap.add_argument("--landmarks", choices=["center", "centroid", "auto"], default="center",
help="center/centroid = placeholder; auto = anatomical fiducials "
"(oriented by the DICOM header; see auto_landmarks.py / AUTO_LANDMARKS.md)")
ap.add_argument("--landmarks-csv", default=None, help="known points to fill in")
ap.add_argument("--min-slices", type=int, default=3,
help="ignore series with fewer slices than this")
ap.add_argument("--relative", action="store_true",
help="write relative paths in filename lists (default: absolute)")
ap.add_argument("--overwrite", action="store_true",
help="reconvert volumes even if the .nii.gz already exists")
args = ap.parse_args()
root = os.path.abspath(args.data)
img_out = os.path.join(root, args.images_dir)
fn_out = os.path.join(root, "filenames")
lm_out = os.path.join(root, "landmarks")
for d in (img_out, fn_out, lm_out):
os.makedirs(d, exist_ok=True)
backend = "SimpleITK" if _HAVE_SITK else ("pydicom" if _HAVE_PYDICOM else "mini_dicom (built-in)")
print(f"[prepare_data] backend: {backend}")
print(f"[prepare_data] scanning: {root}")
csv_points = load_csv(args.landmarks_csv)
if args.num_landmarks < 15:
print("[warn] num-landmarks < 15; the code indexes landmark 13/14 (AC/PC). Forcing 15.")
args.num_landmarks = 15
used_names = set()
image_lines, landmark_lines = [], []
n_series = 0
for dirpath, dcm_files in find_dicom_dirs(root, args.images_dir):
if _HAVE_SITK:
groups = group_series_sitk(dirpath) or group_series_fallback(dcm_files)
else:
groups = group_series_fallback(dcm_files)
for series_key, files in groups.items():
if len(files) < args.min_slices:
continue
name = sanitize(root, dirpath, series_key, used_names)
nii_path = os.path.join(img_out, name + ".nii.gz")
try:
if args.overwrite or not os.path.exists(nii_path):
vol, spacing = load_series(files, dirpath)
if not _HAVE_SITK:
write_nii(nii_path, vol, spacing)
else:
reader = sitk.ImageSeriesReader(); reader.SetFileNames(files)
sitk.WriteImage(reader.Execute(), nii_path)
vol, spacing = load_series(files, dirpath) # for landmark shape
else:
from read_nifti import read_nii
vol, spacing = read_nii(nii_path)
if vol.ndim == 4:
vol = vol[..., 0]
except Exception as e:
print(f"[skip] {dirpath} [{series_key[:12]}]: {e}")
continue
# landmark file
if args.landmarks == "auto":
from auto_landmarks import anatomical_fiducials
cos = _series_cosines(files)
_, lm = anatomical_fiducials(vol, cosines=cos, spacing=spacing,
n=args.num_landmarks)
# let CSV points override specific indices if provided
for idx, xyz in (csv_points.get(name) or {}).items():
if 0 <= idx < len(lm):
lm[idx] = np.clip(np.round(xyz), 0, np.array(vol.shape) - 1).astype(int)
else:
lm = make_landmarks(vol, args.landmarks, args.num_landmarks,
csv_points.get(name))
lm_path = os.path.join(lm_out, name + ".txt")
with open(lm_path, "w") as f:
f.write("\n".join(f"{int(x)},{int(y)},{int(z)}" for x, y, z in lm) + "\n")
ip = nii_path if not args.relative else os.path.relpath(nii_path, root)
lp = lm_path if not args.relative else os.path.relpath(lm_path, root)
image_lines.append(ip)
landmark_lines.append(lp)
n_series += 1
print(f" + {name} shape={tuple(vol.shape)} slices={len(files)}")
with open(os.path.join(fn_out, "image_files.txt"), "w") as f:
f.write("\n".join(image_lines) + ("\n" if image_lines else ""))
with open(os.path.join(fn_out, "landmark_files.txt"), "w") as f:
f.write("\n".join(landmark_lines) + ("\n" if landmark_lines else ""))
print(f"\n[prepare_data] done: {n_series} volume(s)")
print(f" images -> {img_out}")
print(f" filenames -> {os.path.join(fn_out, 'image_files.txt')} (+ landmark_files.txt)")
print(f" landmarks -> {lm_out}")
have_real = bool(csv_points)
if not have_real:
print("\n" + "!" * 72)
print("! LANDMARKS ARE PLACEHOLDERS (image center/centroid), NOT real anatomy.")
print("! They let the pipeline + physics run, but you MUST replace them with")
print("! real annotations (or pass --landmarks-csv) before training a detector.")
print("!" * 72)
if __name__ == "__main__":
main()
|