{
  "fetched_at": "2026-10-20T05:31:07Z",
  "source_url": "https://www.kaggle.com/competitions/bau-ain3002-spring23-a3/data",
  "pages": [
    {
      "name": "data-description",
      "content": "The dataset consists of pathology reports and corresponding Fuhrman grades of renal cell carcinoma patients in The Cancer Genome Atlas. The dataset is modified from the dataset provided in [1] by Hakan Kıymaz (graduate student, BAU). The labels (Fuhrman grades: G1, G2, G3, or G4) for the training set samples are provided.\n\n## Example from the dataset\n\n**Pathology report text:**\n```\nprocedure: right radical nephrectomy tumor type: papillary renal cell carcinoma nuclear grade: grade iii tumor size: 12 x 11.5 x 9 cm renal vein invasion: not present lymph nodes: not examined phthologic stage: t2 nx mx\n```\n\n**Fuhrman grade:**\n```\nG3\n```\n\nThe file ``data.tsv`` provides the list of patients and corresponding pathology reports. The file ``train.csv`` provides the list of patients and corresponding Fuhrman grades as one-hot vectors in the training set. The file ``test.csv`` provides the list of patients in the test set.\n\n## Files\n\n* **data.tsv** - all pathology reports\n     * *pateint_id* - patient id\n     * *pathology_report_text* - unstructured text inside a pathology report\n* **train.csv** - the training set\n     * *patient_id* - patient id\n     * *G1* - the value of class G1 bit in the one-hot vector\n     * *G2* - the value of class G2 bit in the one-hot vector\n     * *G3* - the value of class G3 bit in the one-hot vector\n     * *G4* - the value of class G4 bit in the one-hot vector\n*   **test.csv** - the test set\n     * *patient_id* - patient id\n*   **sample_submission.csv** - a sample submission file in the correct format\n     * *patient_id* - patient id\n     * *likelihood_G1* - a sample's predicted likelihood of belonging to the class G1\n     * *likelihood_G2* - a sample's predicted likelihood of belonging to the class G2\n     * *likelihood_G3* - a sample's predicted likelihood of belonging to the class G3\n     * *likelihood_G4* - a sample's predicted likelihood of belonging to the class G4\n\n[1] Wu, J., Tang, K., Zhang, H., Wang, C., & Li, C. (2020, December). Structured information extraction of pathology reports with attention-based graph convolutional network. In *2020 IEEE International Conference on Bioinformatics and Biomedicine (BIBM)* (pp. 2395-2402). IEEE.",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Description",
      "content": "Cancer grading is important for therapeutic decision making and prognosis prediction. Pathologists grade tumors based on the morphology of cells within tissue and state in the pathology reports in addition to some other clinical parameters, like stage, size of the tumor, and anatomical position. In retrospective studies, extracting these information from unstructured text in pathology reports is crucial for sample selection and experiment design.\n\n**Aim:** To design deep learning models extracting Fuhrman grade from pathology reports of renal cell carcinoma patients  in [The Cancer Genome Atlas (TCGA)](https://www.cancer.gov/about-nci/organization/ccg/research/structural-genomics/tcga).\n\n**Dataset:** Consisting of pathology reports and corresponding Fuhrman grades of renal cell carcinoma patients in The Cancer Genome Atlas. The dataset is modified from the dataset provided in [1].\n\n**Task:** Given text in a patient's pathology report, extract the Fuhrman grade of renal cell carcinoma.\n\n\n## Example from the dataset\n\n**Pathology report text:**\n```\nprocedure: right radical nephrectomy tumor type: papillary renal cell carcinoma nuclear grade: grade iii tumor size: 12 x 11.5 x 9 cm renal vein invasion: not present lymph nodes: not examined phthologic stage: t2 nx mx\n```\n\n**Fuhrman grade:**\n```\nG3\n```\n\n## Acknowledgements\n\nWe thank Hakan Kıymaz (graduate student, BAU) for preparing the competition's dataset from the one published in [1]\n\n[1] Wu, J., Tang, K., Zhang, H., Wang, C., & Li, C. (2020, December). Structured information extraction of pathology reports with attention-based graph convolutional network. In *2020 IEEE International Conference on Bioinformatics and Biomedicine (BIBM)* (pp. 2395-2402). IEEE.",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "Evaluation",
      "content": "The evaluation metric for this competition is negative log likelihood loss:\n\n$$ L = -\\frac{1}{N} \\sum\\_{i=1}^N \\sum\\_{j=1}^K y\\_{i,j}\\log(\\hat{y}\\_{ij}) $$\n\nwhere `N` is the number of samples, `K` is the number of classes, \\\\(y\\_{i,j}\\\\) is 1 if the sample `i` belongs to the class `j`, \\\\(\\hat{y}\\_{ij}\\\\) is the sample `i`'s predicted likelihood of belonging to the class `j`, and \\\\(\\log\\\\) is the natural logarithm.\n\n## Submission Format\n\n**For every pathology report in the dataset**, submission files should contain five columns: `patient_id`, `likelihood_G1`, `likelihood_G2`, `likelihood_G3`, and `likelihood_G4`. Please see the `sample_submission.csv` for the submission file format.",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    },
    {
      "name": "rules",
      "content": "Enjoy the competition!",
      "isPublished": false,
      "hasIsPublished": false,
      "postTitle": "",
      "hasPostTitle": false,
      "mimeType": "",
      "hasMimeType": false
    }
  ]
}