File size: 15,679 Bytes
e2f75d1 | 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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 | #!/usr/bin/env python3
"""
IDRiD adapter — DR grading + lesion segmentation + OD/fovea localization (single cohort).
Two image tracks (disjoint, distinguished by basename prefix):
Grading + Localization track (516 = 413 train + 103 test):
Inputs:
B. Disease Grading/1. Original Images/{a|b}. {Training|Testing} Set/IDRiD_NNN.jpg
B. Disease Grading/2. Groundtruths/{a|b}. IDRiD_Disease Grading_{Training|Testing} Labels.csv
C. Localization/2. Groundtruths/1. Optic Disc Center Location/*.csv
C. Localization/2. Groundtruths/2. Fovea Center Location/*.csv
Manifest row: DR grade + DME risk; has_segmentation=False.
Sidecar parquet: OD + fovea pixel coords + image dims.
Segmentation track (81 = 54 train + 27 test):
Inputs:
A. Segmentation/1. Original Images/{a|b}. {Training|Testing} Set/IDRiD_NN.jpg
A. Segmentation/2. All Segmentation Groundtruths/.../{MA,HE,EX,SE,OD}.tif
Mask suffix per class: MA=microaneurysms, HE=haemorrhages, EX=hard exudates,
SE=soft exudates, OD=optic disc.
Manifest row: has_segmentation=True; lesion_tags driven by which masks exist.
Outputs (under {output_root}):
extracted/public_idrid/{hash[:2]}/{hash}/
fundus_color.jpg
lesion_microaneurysms.png (binary 0/255, present only in seg track)
lesion_haemorrhages.png
lesion_hard_exudates.png
lesion_soft_exudates.png
optic_disc_mask.png
meta.json
manifest/public_idrid_images.parquet
manifest/public_idrid_sidecar.parquet (localization coords)
captions/public_idrid_captions.parquet
"""
import argparse
import json
from pathlib import Path
import numpy as np
import pandas as pd
from PIL import Image
from public_common import (
IMAGE_SCHEMA_COLUMNS, CAPTION_SCHEMA_COLUMNS,
study_hash_for, default_base_fields,
caption_l1_public, caption_l3_public,
study_dir_for, rel_file_path, write_meta, coerce_image_row,
)
COHORT = "public_idrid"
COHORT_PHRASE = "IDRiD diabetic retinopathy dataset"
DR_SEVERITY = {0: "none", 1: "mild", 2: "moderate", 3: "severe", 4: "proliferative"}
DME_RISK = {0: None, 1: "macular_edema", 2: "clinically_significant_macular_edema"}
LESION_SUFFIX = {
"MA": ("microaneurysms", "lesion_microaneurysms.png"),
"HE": ("haemorrhages", "lesion_haemorrhages.png"),
"EX": ("hard_exudates", "lesion_hard_exudates.png"),
"SE": ("soft_exudates", "lesion_soft_exudates.png"),
"OD": ("optic_disc", "optic_disc_mask.png"),
}
def _binarize_save(src: Path, dst: Path):
arr = np.array(Image.open(src).convert("L"))
Image.fromarray(((arr > 0).astype(np.uint8) * 255), mode="L").save(
dst, "PNG", optimize=True)
def _save_fundus(src: Path, dst: Path):
img = Image.open(src).convert("RGB")
img.save(dst, "JPEG", quality=95)
return img.size # (w, h)
# ============================================================
# Grading + Localization track
# ============================================================
def _load_grading_labels(in_root: Path) -> pd.DataFrame:
parts = []
for split, fn in [
("train", "B. Disease Grading/2. Groundtruths/a. IDRiD_Disease Grading_Training Labels.csv"),
("test", "B. Disease Grading/2. Groundtruths/b. IDRiD_Disease Grading_Testing Labels.csv"),
]:
df = pd.read_csv(in_root / fn)
df = df[[c for c in df.columns if not c.startswith("Unnamed")]]
df.columns = [c.strip() for c in df.columns]
df = df.rename(columns={
"Image name": "image_name",
"Retinopathy grade": "dr_grade",
"Risk of macular edema": "dme_risk",
})
df = df.dropna(subset=["image_name"])
df["split"] = split
parts.append(df)
return pd.concat(parts, ignore_index=True)
def _load_loc_csv(p: Path, prefix: str) -> dict:
df = pd.read_csv(p)
df = df[[c for c in df.columns if not c.startswith("Unnamed")]]
df.columns = [c.strip() for c in df.columns]
df = df.rename(columns={"Image No": "image_name",
"X- Coordinate": f"{prefix}_x",
"Y - Coordinate": f"{prefix}_y"})
df = df.dropna(subset=["image_name"])
return df.set_index("image_name")[[f"{prefix}_x", f"{prefix}_y"]].to_dict(orient="index")
def process_grading(in_root: Path, out_root: Path, force: bool):
labels = _load_grading_labels(in_root)
od_train = _load_loc_csv(in_root / "C. Localization/2. Groundtruths/1. Optic Disc Center Location/a. IDRiD_OD_Center_Training Set_Markups.csv", "od")
od_test = _load_loc_csv(in_root / "C. Localization/2. Groundtruths/1. Optic Disc Center Location/b. IDRiD_OD_Center_Testing Set_Markups.csv", "od")
fv_train = _load_loc_csv(in_root / "C. Localization/2. Groundtruths/2. Fovea Center Location/IDRiD_Fovea_Center_Training Set_Markups.csv", "fovea")
fv_test = _load_loc_csv(in_root / "C. Localization/2. Groundtruths/2. Fovea Center Location/IDRiD_Fovea_Center_Testing Set_Markups.csv", "fovea")
od_all = {**od_train, **od_test}
fv_all = {**fv_train, **fv_test}
rows, caps, side = [], [], []
for _, lab in labels.iterrows():
name = str(lab["image_name"]).strip()
split = lab["split"]
src = in_root / f"B. Disease Grading/1. Original Images/{'a. Training Set' if split=='train' else 'b. Testing Set'}/{name}.jpg"
if not src.exists():
print(f" [grading] missing image {src}")
continue
basename = f"grading_{split}_{name}"
sh = study_hash_for(COHORT, basename)
sdir = study_dir_for(out_root, COHORT, sh)
sdir.mkdir(parents=True, exist_ok=True)
meta_p = sdir / "meta.json"
if meta_p.exists() and not force:
try:
meta = json.loads(meta_p.read_text())
if meta.get("status") == "ok":
rows.append(_grading_row(meta))
caps.extend(_grading_caps(meta))
side.append(_sidecar_row(meta))
continue
except Exception:
pass
w, h = _save_fundus(src, sdir / "fundus_color.jpg")
od = od_all.get(name, {})
fv = fv_all.get(name, {})
meta = {
"status": "ok", "cohort": COHORT, "study_hash": sh,
"source_basename": basename, "track": "grading",
"split": split,
"image_height_px": int(h), "image_width_px": int(w),
"eye": "unknown",
"dr_grade": int(lab["dr_grade"]) if pd.notna(lab["dr_grade"]) else None,
"dme_risk": int(lab["dme_risk"]) if pd.notna(lab["dme_risk"]) else None,
"od_x_px": float(od.get("od_x")) if od else None,
"od_y_px": float(od.get("od_y")) if od else None,
"fovea_x_px": float(fv.get("fovea_x")) if fv else None,
"fovea_y_px": float(fv.get("fovea_y")) if fv else None,
}
write_meta(sdir, meta)
rows.append(_grading_row(meta))
caps.extend(_grading_caps(meta))
side.append(_sidecar_row(meta))
print(f"[{COHORT}/grading] {len(rows)} rows")
return rows, caps, side
def _grading_row(meta: dict) -> dict:
sh = meta["study_hash"]
image_id = f"{COHORT}_{sh}_fundus_color"
row = default_base_fields(COHORT, sh, eye=meta["eye"])
dr = meta.get("dr_grade")
dme_risk = meta.get("dme_risk")
dx, lesions = [], []
if dr is not None and dr > 0:
dx.append("DR")
if dme_risk and dme_risk > 0:
dx.append("DME")
tag = DME_RISK.get(dme_risk)
if tag:
lesions.append(tag)
row["diagnosis_group"] = dx
row["lesion_tags"] = lesions
row["severity"] = DR_SEVERITY.get(dr, "unknown")
row["diagnosis_source"] = "expert_grade" if dr is not None else "none"
row["label_confidence"] = "consensus" if dr is not None else None
row.update({
"image_id": image_id,
"file_path": rel_file_path(COHORT, sh, "fundus_color.jpg"),
"file_format": "jpg",
"modality": "fundus_color", "anatomy": "macula",
"device_technology": "fundus_camera", "scan_protocol": "single_shot",
"image_height_px": meta["image_height_px"],
"image_width_px": meta["image_width_px"],
"has_segmentation": False, "n_layers_visible": 0,
"is_valid": True,
})
return row
def _grading_caps(meta: dict) -> list:
sh = meta["study_hash"]
image_id = f"{COHORT}_{sh}_fundus_color"
caps = caption_l1_public(image_id, COHORT_PHRASE, "fundus_color", meta["eye"])
parts = [f"A color fundus photograph from the IDRiD dataset ({meta['split']} split)"]
if meta.get("dr_grade") is not None:
parts.append(f"diabetic retinopathy grade {meta['dr_grade']} ({DR_SEVERITY[meta['dr_grade']]})")
if meta.get("dme_risk") is not None and meta["dme_risk"] > 0:
parts.append(f"macular edema risk {meta['dme_risk']}")
l3 = ", ".join(parts) + "."
caps.append(caption_l3_public(image_id, l3, "manifest_fields+csv_labels"))
return caps
def _sidecar_row(meta: dict) -> dict:
sh = meta["study_hash"]
return {
"image_id": f"{COHORT}_{sh}_fundus_color",
"split": meta.get("split"),
"od_x_px": meta.get("od_x_px"),
"od_y_px": meta.get("od_y_px"),
"fovea_x_px": meta.get("fovea_x_px"),
"fovea_y_px": meta.get("fovea_y_px"),
"image_width_px": meta.get("image_width_px"),
"image_height_px": meta.get("image_height_px"),
}
# ============================================================
# Segmentation track
# ============================================================
def process_segmentation(in_root: Path, out_root: Path, force: bool):
seg_dir = in_root / "A. Segmentation"
rows, caps = [], []
for split, sub_img, sub_gt in [
("train", "a. Training Set", "a. Training Set"),
("test", "b. Testing Set", "b. Testing Set"),
]:
img_dir = seg_dir / "1. Original Images" / sub_img
gt_dir = seg_dir / "2. All Segmentation Groundtruths" / sub_gt
for ip in sorted(img_dir.glob("IDRiD_*.jpg")):
name = ip.stem # IDRiD_01
basename = f"seg_{split}_{name}"
sh = study_hash_for(COHORT, basename)
sdir = study_dir_for(out_root, COHORT, sh)
sdir.mkdir(parents=True, exist_ok=True)
meta_p = sdir / "meta.json"
if meta_p.exists() and not force:
try:
meta = json.loads(meta_p.read_text())
if meta.get("status") == "ok":
rows.append(_seg_row(meta))
caps.extend(_seg_caps(meta))
continue
except Exception:
pass
w, h = _save_fundus(ip, sdir / "fundus_color.jpg")
has_mask = {}
for suffix, (label, fname) in LESION_SUFFIX.items():
subdir_idx = {"MA": "1. Microaneurysms", "HE": "2. Haemorrhages",
"EX": "3. Hard Exudates", "SE": "4. Soft Exudates",
"OD": "5. Optic Disc"}[suffix]
src = gt_dir / subdir_idx / f"{name}_{suffix}.tif"
if src.exists():
_binarize_save(src, sdir / fname)
has_mask[label] = True
else:
has_mask[label] = False
meta = {
"status": "ok", "cohort": COHORT, "study_hash": sh,
"source_basename": basename, "track": "segmentation",
"split": split,
"image_height_px": int(h), "image_width_px": int(w),
"eye": "unknown",
**{f"has_{k}_mask": v for k, v in has_mask.items()},
}
write_meta(sdir, meta)
rows.append(_seg_row(meta))
caps.extend(_seg_caps(meta))
print(f"[{COHORT}/seg] {len(rows)} rows")
return rows, caps
def _seg_row(meta: dict) -> dict:
sh = meta["study_hash"]
image_id = f"{COHORT}_{sh}_fundus_color"
row = default_base_fields(COHORT, sh, eye=meta["eye"])
lesions = []
for label in ("microaneurysms", "haemorrhages", "hard_exudates", "soft_exudates"):
if meta.get(f"has_{label}_mask"):
lesions.append(label)
if lesions:
row["diagnosis_group"] = ["DR"] # any DR-related lesion implies DR positive
row["lesion_tags"] = lesions
row["diagnosis_source"] = "expert_segmentation" if lesions else "none"
row.update({
"image_id": image_id,
"file_path": rel_file_path(COHORT, sh, "fundus_color.jpg"),
"file_format": "jpg",
"modality": "fundus_color", "anatomy": "macula",
"device_technology": "fundus_camera", "scan_protocol": "single_shot",
"image_height_px": meta["image_height_px"],
"image_width_px": meta["image_width_px"],
"has_segmentation": True, "n_layers_visible": 0,
"is_valid": True,
})
return row
def _seg_caps(meta: dict) -> list:
sh = meta["study_hash"]
image_id = f"{COHORT}_{sh}_fundus_color"
caps = caption_l1_public(image_id, COHORT_PHRASE, "fundus_color", meta["eye"])
present = [k for k in ("microaneurysms", "haemorrhages", "hard_exudates", "soft_exudates", "optic_disc")
if meta.get(f"has_{k}_mask")]
parts = [f"A color fundus photograph from the IDRiD dataset ({meta['split']} split, segmentation subset)"]
if present:
parts.append("with manually annotated " + ", ".join(p.replace("_", " ") for p in present) + " segmentation masks")
l3 = ", ".join(parts) + "."
caps.append(caption_l3_public(image_id, l3, "manifest_fields+mask_presence"))
return caps
# ============================================================
# Main
# ============================================================
def main():
ap = argparse.ArgumentParser()
ap.add_argument("--input-root", required=True,
help="Path to .../Generation/IDRiD (contains A./B./C. subdirs)")
ap.add_argument("--output-root", required=True)
ap.add_argument("--force", action="store_true")
ap.add_argument("--skip-grading", action="store_true")
ap.add_argument("--skip-segmentation", action="store_true")
args = ap.parse_args()
in_root = Path(args.input_root)
out_root = Path(args.output_root)
rows, caps, side = [], [], []
if not args.skip_grading:
r, c, s = process_grading(in_root, out_root, args.force)
rows += r; caps += c; side += s
if not args.skip_segmentation:
r, c = process_segmentation(in_root, out_root, args.force)
rows += r; caps += c
if not rows:
print(f"[{COHORT}] no rows produced")
return
mdir = out_root / "manifest"
cdir = out_root / "captions"
mdir.mkdir(parents=True, exist_ok=True)
cdir.mkdir(parents=True, exist_ok=True)
imgs_df = pd.DataFrame([coerce_image_row(r) for r in rows])[IMAGE_SCHEMA_COLUMNS]
imgs_df.to_parquet(mdir / f"{COHORT}_images.parquet", index=False)
caps_df = pd.DataFrame(caps)[CAPTION_SCHEMA_COLUMNS]
caps_df.to_parquet(cdir / f"{COHORT}_captions.parquet", index=False)
if side:
pd.DataFrame(side).to_parquet(mdir / f"{COHORT}_sidecar.parquet", index=False)
print(f"[{COHORT}] wrote {len(imgs_df)} images, {len(caps_df)} captions, {len(side)} sidecar")
print(imgs_df.groupby(["has_segmentation", "severity"]).size().to_string())
if __name__ == "__main__":
main()
|