Kidney-CT-Abnormality

This Kidney-CT-Abnormality dataset consists of kidney CT scans with abnormality label.

Dataset Details

Dataset Description

This Kidney-CT-Abnormality dataset comprises a comprehensive collection CT scans focusing on kidney CT abnormality, which can serve as a resource for researchers with this field.

Contained within this dataset are 986 .mha (medical high-resolution image) files, which are all 3D medical iamges. 3D images means multiple layers are included in each image, which can be beneficial for precise classification. A .json file is also included, illustrating the abnormality status of each image.

Note that, as stated by the authors, this dataset was reconstructed from “Dataset for: Kidney abnormality segmentation in thorax-abdomen CT scans” (https://zenodo.org/records/8014290), to fit the 2023 automated universal classification challenges (AUC2023).

In a nutshell, the Kidney-CT-Abnormality dataset can potentially can serve for the academic and research community, possibly enhancing studies in medical image processing and diagnostic algorithm development, thereby improving understanding of kidney diseases and diagnostic accuracy through technological advancements.

Dataset Sources

• Original Homepage: https://zenodo.org/records/8043408

Uses

This dataset is intended for kidney abnormality classification.

Direct Use

By loading this dataset, it can output transformed images (mnumpy array dtype=uint32, and have broken down to sequence of images rather than multiple layer images), the original image path (numpy array dtype=float64 after loading), and the image label. Along with these information, various classification task can be performed.

Out-of-Scope Use

This dataset cannot be utilized for segmentation task since no ground truth image included.

Dataset Initial Processing

• Train-Test soplit: The train test split were created to fit for further classification task. Images are distributed to train/test folder randomly.
• Metadata modification: The original json file includes other information like brief descriptions and license. Only image file name, corresponding labels (Normal: 0, Abnormal: 1) were preserved, and train-test split information were added. Note that the original file name in the json file did not match the image file name (format issue), thus also gone through modificaitons.

Dataset Structure

Image data ()

• kidney_CT
  • kidney_CT.zip
    • train
      • kidneyabnormalityKiTS-xxxx_0000.mha (The 'KiTs' can be replaced by 'RUMC', the 'xxxx' is a four-digits number, serving as image number)
      • ...
    • test
      • kidneyabnormalityKiTS-xxxx_0000.mha (The 'KiTs' can be replaced by 'RUMC', the 'xxxx' is a four-digits number, serving as image number)
      • ...

Metadata (annotations indicating the abnormality status of the image, and whether sperate to train or test group)

• dataset_m.json
  • {"image":"kidneyabnormalityKiTS-0000_0000.mha","split":"train","abnormality":0}
  • ...

Dataset Creation

Curation Rationale

Kidney diseases often present significant detection challenges, yet their timely identification and diagnosis are critical for effective treatment.

The Kidney-CT-Abnormality dataset is curated with the express purpose of facilitating the development of advanced artificial intelligence (AI) and machine learning algorithms aimed at enhancing the recognition and diagnostic accuracy of kidney diseases.

Source Data

Data Collection and Processing

This dataset's original homepage: https://zenodo.org/records/8043408 The dataset was adapted from https://zenodo.org/records/8014290 as mentioned before. The original dataset contains “215 thoraxabdomen CT scans with segmentations of the kidney and abnormalities in the kidney”. Note that the original datasets contains in total 38.4G image data in mha format, and there is no .json file indicating the abnormality status. Alternatively, a segmentation kidney image dataset is included.

Annotations

The annotation information, which are the abnormality labels, are completed by the original authors and included the json file.

Personal and Sensitive Information

The CT scan image file names contain the study ID, indicating the differences in scans. These ids are anonymous and don’t contain any other personal information.

Some helper functions for further utilization

Bias, Risks, and Limitations

The dataset only include abnormality label, with no further implication of specific diseases. This can limit the algorithms diagnostic specificity. Moreover, the collected data can have potential bias. For instance, the CT scans might be generated from specific demographics, which can introduce bias (skewing the representation and applicability of the data).

Recommendations

Users should be made aware of the risks, biases and limitations of the dataset. More information needed for further recommendations.

Citation

Datasets: Alves, N., & Boulogne, L. (2023). Kidney CT Abnormality [Data set]. Zenodo. https://doi.org/10.5281/zenodo.8043408 Gabriel E. Humpire-Mamani, Luc Builtjes, Colin Jacobs, Bram van Ginneken, Mathias Prokop, & Ernst Th. Scholten. (2023). Dataset for: Kidney abnormality segmentation in thorax-abdomen CT scans [Data set]. Zenodo. https://doi.org/10.5281/zenodo.8014290