prostate158.py

import datasets
from typing import List

import pandas as pd
import numpy as np
from pathlib import Path
import os

_DESCRIPTION = "Prostate dataset."

class Prostate158Dataset(datasets.GeneratorBasedBuilder):
    VERSION = datasets.Version("1.1.0")

    BUILDER_CONFIGS = [
        datasets.BuilderConfig(
            name="2d",
            version=VERSION,
            description="Return all the dataset in a 2d image format",
        ),
        # datasets.BuilderConfig(
        #     name="2d_array",
        #     version=VERSION,
        #     description="Return all the dataset in a 2d array format",
        # ),
        # datasets.BuilderConfig(
        #     name="3d_array",
        #     version=VERSION,
        #     description="Return all the dataset in a 2d array format",
        # ),
        datasets.BuilderConfig(
            name="3d_path",
            version=VERSION,
            description="Return all the dataset in a 3d path format",
        ),
        
    ]

    DEFAULT_CONFIG_NAME = "3d_path"

    def _info(self):
        if self.config.name == "2d":
            features = datasets.Features(
                {
                    "t2": datasets.Image(),
                    "adc": datasets.Image(),
                    "dwi": datasets.Image(),
                    "t2_anatomy_reader1": datasets.Image(),
                    "adc_tumor_reader1": datasets.Image(),
                }
            )
        # elif self.config.name == "2d_array":
        #     features = datasets.Features(
        #         {
        #             "t2": datasets.Array2D(shape=(200,200), dtype="int"),
        #             "adc": datasets.Array2D(shape=(200,200), dtype="int"),
        #             "dwi": datasets.Array2D(shape=(200,200), dtype="int"),
        #             "t2_anatomy_reader1": datasets.Array2D(shape=(200,200), dtype="int"),
        #             "adc_tumor_reader1": datasets.Array2D(shape=(200,200), dtype="int"),
        #         }
        #     )
            
        elif self.config.name == "3d_path":
            features = datasets.Features(
                {
                    "t2_path": datasets.Value(dtype="string"),
                    "adc_path": datasets.Value(dtype="string"),
                    "dwi_path": datasets.Value(dtype="string"),
                    "t2_anatomy_reader1_path": datasets.Value(dtype="string"),
                    "adc_tumor_reader1_path": datasets.Value(dtype="string"),
                }
            )

        # elif self.config.name == "3d_array":
        #     features = datasets.Features(
        #         {
        #             "t2": datasets.Array3D(shape=(200,200,20), dtype="int"),
        #             "adc": datasets.Array3D(shape=(200,200,20), dtype="int"),
        #             "dwi": datasets.Array3D(shape=(200,200,20), dtype="int"),
        #             "t2_anatomy_reader1": datasets.Array3D(shape=(200,200,20), dtype="int"),
        #             "adc_tumor_reader1": datasets.Array3D(shape=(200,200,20), dtype="int"),
        #         }
        #     )
    
        return datasets.DatasetInfo(
            description=_DESCRIPTION,
            features=features,
        )

    def _split_generators(
        self,
        dl_manager: datasets.DownloadManager
    ) -> List[datasets.SplitGenerator]:
        downloaded_files = dl_manager.download_and_extract("data.zip")

        return [
            datasets.SplitGenerator(
                name=datasets.Split.TRAIN,
                gen_kwargs={
                    "split": "train",
                    "downloaded_files": Path(downloaded_files),
                },
            ),
            datasets.SplitGenerator(
                name=datasets.Split.VALIDATION,
                gen_kwargs={
                    "split": "valid",
                    "downloaded_files": Path(downloaded_files),
                },
            ),
            datasets.SplitGenerator(
                name=datasets.Split.TEST,
                gen_kwargs={
                    "split": "test",
                    "downloaded_files": Path(downloaded_files),
                },
            ),
        ]
    
    def _generate_examples(self, split, downloaded_files):
        images_list = ["t2", "adc", "dwi", "t2_anatomy_reader1", "adc_tumor_reader1"]
        df = pd.read_csv(downloaded_files / f"{split}.csv")
        if self.config.name == "2d":
            import nibabel as nib
            from PIL import Image # moving imports here to not require unnecessary packages to other users
            
            yield_index = -1
            for row in df.to_dict(orient="records"):
                images_data = {image_name: nib.load(downloaded_files / row[image_name]).get_fdata() for image_name in images_list}
                for i in range(images_data["t2"].shape[2]):
                    yield_index += 1
                    yield yield_index, {image_name: Image.fromarray(image[:, :, i]) for image_name, image in images_data.items()}

        # elif self.config.name == "2d_array":
        #     import nibabel as nib
            
        #     yield_index = -1
        #     for row in df.to_dict(orient="records"):
        #         images_data = {image_name: nib.load(downloaded_files / row[image_name]).get_fdata() for image_name in images_list}
        #         for i in range(images_data["t2"].shape[2]):
        #             yield_index += 1
        #             yield yield_index, {image_name: image[:, :, i] for image_name, image in images_data.items()}

        elif self.config.name == "3d_path":
            for idx, row in enumerate(df.to_dict(orient="records")):
                yield idx, {image_name+"_path": downloaded_files / row[image_name] for image_name in images_list}

        # elif self.config.name == "3d_array":
        #     import nibabel as nib
            
        #     for idx, row in enumerate(df.to_dict(orient="records")):
        #         yield idx, {image_name: nib.load(downloaded_files / row[image_name]).get_fdata() for image_name in images_list}