KBSMC_gastric_cancer_grading_dataset

This repository provides the KBSMC gastric cancer grading dataset that has been introduced in the paper:

Download

Google drive:
Training + Validation + Testing (Resize to 512): [link]

Brief description

The tissue images and annotations are provided by Kangbuk Samsung Hospital, Seoul, South Korea.
Two pathologists have delineated the annotation: Kim, Kyungeun, and Song, Boram.
Herein, we obtained the benign (BN) and three cancer ROIs, including tubular well-differentiated adenocarcinoma (TW), tubular moderately-differentiated adenocarcinoma (TM), and tubular poorly-differentiated adenocarcinoma (TP) tumor.

• The train+valid+test sets contain 98 whole slide images (WSIs) from 98 patients that were collected between 2016 and 2020 from Kangbuk Samsung Hospital (IRB No. 2021-04-035) and scanned at 40x magnification using an Aperio digital slide scanner (Leica Biosystems).

• The patch images are generated at 40x of size (~270 µm x 0.270 µm), then resize to 512x512 pixels (20x). For more detail, please refer to the paper above.

Dataset detail

Status	Training	Validation	Testing
Benign	20,883	8,398	7,955
TW	14,251	2,239	1,795
TM	20,815	2,370	2,458
TP	27,689	2,374	3,579

Training + Validation + Testing Structure

KBSMC_gastric_cancer_grading_512
├── WSIs
│ ├── WSIs_001
│ │ ├── patch_1152_6592_class_2.jpg
│ │ ├── patch_1344_6720_class_2.jpg
│ │ ├── patch_1344_6848_class_2.jpg
│ │ ├── ...
│ ├── WSIs_002
│ ├── WSIs_003
│ ├── WSIs_004
│ ├── ...
│ └── WSIs_158
└── WSIs_Split.csv

Notes:

Train+Valid+Test The class labels are determined by the last digit in the image name (bolded), and there are 5 digits from 1 to 5:

1:"BN" 2:"BN", 3:"TW", 4:"TM", 5:"TP"

We consolidate 1 and 2. Specifically, when the last digit is 1 or 2, the class is considered "Benign."

For example, if the image name is "patch_XXXX_YYYY_class_1.jpg," it belongs to the benign class. This categorization aligns with the usage of four datasets in our paper.

Please refer to the data loading code for more details.

Simple way to load the dataset

Check out the dataset.py

def prepare_gastric_data(data_root_dir='./KBSMC_Gastric_WSI_Cancer_Grading_1024/', nr_classes=4):
    def load_data_info_from_list(wsi_list, data_root_dir, gt_list, nr_claases):
        file_list = []
        for wsi_name in wsi_list:
            pathname = glob(f'{data_root_dir}/{wsi_name}/*.jpg')
            file_list.extend(pathname)
        label_list = [int(file_path.split('_')[-1].split('.')[0]) for file_path in file_list]
        label_list = [gt_list[i] for i in label_list]
        list_out = list(zip(file_list, label_list))

        list_out = [list_out[i] for i in range(len(list_out)) if list_out[i][1] < nr_claases]
        return list_out

    gt_list = {1: 0,  # "BN", #0
               2: 0,  # "BN", #0
               3: 1,  # "TW", #2
               4: 2,  # "TM", #3
               5: 3,  # "TP", #4
               }

    WSI_dir = data_root_dir + '/WSIs'
    csv_path = data_root_dir + '/WSIs_Split_Info.csv'

    df = pd.read_csv(csv_path).iloc[:, :3]
    train_list = list(df.query('Task == "train"')['WSI'])
    valid_list = list(df.query('Task == "val"')['WSI'])
    test_list = list(df.query('Task == "test"')['WSI'])


    train_set = load_data_info_from_list(train_list, WSI_dir, gt_list, nr_classes)
    valid_set = load_data_info_from_list(valid_list, WSI_dir, gt_list, nr_classes)
    test_set = load_data_info_from_list(test_list, WSI_dir, gt_list, nr_classes)


    print_number_of_sample(train_set, 'Train')
    print_number_of_sample(valid_set, 'Valid')
    print_number_of_sample(test_set, 'Test')

    return train_set, valid_set, test_set

class DatasetSerial(data.Dataset):
    """get image by index
    """
    def __init__(self, pair_list, img_transform=None, target_transform=None, two_crop=False):
        self.pair_list = pair_list

        self.img_transform = img_transform
        self.target_transform = target_transform
        self.num = self.__len__()

    def __len__(self):
        return len(self.pair_list)

    def __getitem__(self, index):
        """
        Args:
            index (int): index
        Returns:
            tuple: (image, index, ...)
        """
        path, target = self.pair_list[index]
        image = pil_loader(path)

        # # image
        if self.img_transform is not None:
            img = self.img_transform(image)
        else:
            img = image

        return img, target

Citation