qupath_roi_splitter_folder.py

from os import listdir
from os.path import isfile, join
import argparse

import cv2
import geojson
import numpy as np
import pandas as pd


def collect_coords(input_coords, feature_index, coord_index=0):
	coords_with_index = []
	for coord in input_coords:
		coords_with_index.append((coord[0], coord[1], feature_index, coord_index))
		coord_index += 1
	return coords_with_index


def collect_roi_coords(input_roi):
	coords = input_roi["geometry"]["coordinates"]

	def process_coords(coord_list):
		if isinstance(coord_list[0], (int, float)):
			return [coord_list]
		elif all(isinstance(c, list) for c in coord_list):
			return coord_list
		else:
			return [coord_list]

	if isinstance(coords[0][0], list):
		return [process_coords(sub_coords) for sub_coords in coords]
	else:
		return [process_coords(coords)]



def split_qupath_roi_folder(roi_folder):
	geojson_files = [f for f in listdir(roi_folder) if isfile(join(roi_folder, f)) and f.endswith(".geojson")]

	for gj_f in geojson_files:
		with open(join(roi_folder, gj_f)) as file:
			geo_json = geojson.load(file)

			# HE dimensions
			dim_plt = [int(geo_json["dim"]["width"]), int(geo_json["dim"]["height"])]
			tma_name = geo_json["name"]

			if "featureNames" in geo_json:
				cell_types = [ct.rsplit(" - ", 1)[-1] for ct in geo_json["featureNames"]]
			else:
				cell_types = ["all"]

			coords_by_cell_type = {ct: [] for ct in cell_types}
			if "all" not in coords_by_cell_type:
				coords_by_cell_type["all"] = []

			for roi in geo_json["features"]:
				feature_coords = collect_roi_coords(roi)

				if args.all or "classification" not in roi["properties"]:
					coords_by_cell_type["all"].append(feature_coords)
				elif "classification" in roi["properties"]:
					cell_type = roi["properties"]["classification"]["name"]
					if cell_type in cell_types:
						coords_by_cell_type[cell_type].append(feature_coords)

			for cell_type, coords_list in coords_by_cell_type.items():
				if coords_list:
					img = np.ones((dim_plt[1], dim_plt[0], 3), dtype="uint8") * 255

					all_coords = []
					for feature in coords_list:
						for polygon in feature:
							# Multiple sub_polygons in LMD data
							for sub_poly in polygon if isinstance(polygon[0][0], list) else [polygon]:
								pts = np.array(sub_poly, dtype=np.float32).reshape(-1, 2)
								pts = pts.astype(np.int32)

								# Get filled pixel coordinates
								if args.fill:
									temp_img = np.ones((dim_plt[1], dim_plt[0]), dtype="uint8") * 255
									cv2.fillPoly(temp_img, [pts], color=0)
									filled_coords = np.column_stack(np.where(temp_img == 0))
									all_coords.extend(filled_coords[:, [1, 0]])  # Swap columns to get (x, y)
									cv2.fillPoly(img, [pts], color=0)
								else:
									cv2.polylines(img, [pts], isClosed=True, color=(0, 0, 0), thickness=1)
									all_coords.extend(pts)

					all_coords = np.array(all_coords)
					coords_df = pd.DataFrame(all_coords, columns=['x', 'y'], dtype=int)
					coords_df.to_csv("{}_{}.txt".format(tma_name, cell_type), sep='\t', index=False)

					# Generate image for visualization if --img is specified
					if args.img:
						cv2.imwrite("{}_{}.png".format(tma_name, cell_type), img)


if __name__ == "__main__":
	parser = argparse.ArgumentParser(description="Split ROI coordinates of QuPath TMA annotation by cell type (classfication) [FOLDER]")
	parser.add_argument("qupath_roi_folder", default=False, help="Input QuPath annotation folder (GeoJSON files)")
	parser.add_argument("--fill", action="store_true", required=False, help="Fill holes in ROI (order of coordinates will be lost)")
	parser.add_argument('--version', action='version', version='%(prog)s 0.3.2')
	parser.add_argument("--all", action="store_true", required=False, help="Extracts all ROIs")
	parser.add_argument("--img", action="store_true", required=False, help="Generates image of ROIs")
	args = parser.parse_args()

	if args.qupath_roi_folder:
		split_qupath_roi_folder(args.qupath_roi_folder)