Update README.md

README.md

@@ -15,18 +15,20 @@ tags:
 - image-classification
 - image-segmentation
 - object-detection
-dataset_summary: '
+- medical
+dataset_summary: >
 
 
 
 
-  This is a [FiftyOne](https://github.com/voxel51/fiftyone) dataset with 642 samples.
+  This is a [FiftyOne](https://github.com/voxel51/fiftyone) dataset with 642
+  samples.
 
 
   ## Installation
 
 
-  If you haven''t already, install FiftyOne:
+  If you haven't already, install FiftyOne:
 
 
   ```bash
@@ -48,9 +50,9 @@ dataset_summary: '
 
   # Load the dataset
 
-  # Note: other available arguments include ''max_samples'', etc
+  # Note: other available arguments include 'max_samples', etc
 
-  dataset = load_from_hub("harpreetsahota/SLAKE")
+  dataset = load_from_hub("Voxel51/SLAKE")
 
 
   # Launch the App
@@ -58,20 +60,15 @@ dataset_summary: '
   session = fo.launch_app(dataset)
 
   ```
-
-  '
+license: cc-by-4.0
 ---
 
 # Dataset Card for SLAKE
 
-<!-- Provide a quick summary of the dataset. -->
-
-
-
-
-
 This is a [FiftyOne](https://github.com/voxel51/fiftyone) dataset with 642 samples.
 
+![image/png](SLAKE.gif)
+
 ## Installation
 
 If you haven't already, install FiftyOne:
@@ -88,141 +85,185 @@ from fiftyone.utils.huggingface import load_from_hub
 
 # Load the dataset
 # Note: other available arguments include 'max_samples', etc
-dataset = load_from_hub("harpreetsahota/SLAKE")
+dataset = load_from_hub("Voxel51/SLAKE")
 
 # Launch the App
 session = fo.launch_app(dataset)
 ```
 
 
-## Dataset Details
+# SLAKE Dataset Processing Overview
 
-### Dataset Description
+See the accompanying code in this repo for details on how the dataset was parsed.
+
+## Original Structure
+
+```
+SLAKE/imgs/xmlab0/
+├── detection.json       # Organ bounding boxes (absolute coordinates)
+├── question.json        # VQA pairs (English + Chinese)
+├── mask_xmlab0.png      # Segmentation mask (pixel values)
+└── source_xmlab0.jpg    # Medical image
+```
+
+## Key Processing Steps
 
-<!-- Provide a longer summary of what this dataset is. -->
+1. **Coordinate Conversion**: Converted absolute bounding boxes to relative coordinates for FiftyOne
+2. **Language Filtering**: Extracted English-only questions while preserving order
+3. **Mask Labeling**: Used mapping file to convert pixel values (70, 135, etc.) to organ names ("Liver", "Left Kidney")
+4. **Data Structuring**: 
+   - Sample-level: location, modality, base_type
+   - Individual Q&A: question_0/answer_0, question_1/answer_1, etc.
 
+## Final FiftyOne Dataset
+Each sample contains:
+- Medical image with object detections (organ bounding boxes)
+- Multi-class segmentation with readable organ labels  
+- VQA pairs as individual fields
+- Sample metadata (imaging modality, body region, etc.)
 
 
-- **Curated by:** [More Information Needed]
-- **Funded by [optional]:** [More Information Needed]
-- **Shared by [optional]:** [More Information Needed]
-- **Language(s) (NLP):** en
-- **License:** [More Information Needed]
+## Dataset Details
 
-### Dataset Sources [optional]
+### Dataset Description
 
-<!-- Provide the basic links for the dataset. -->
+SLAKE (Semantically-Labeled Knowledge-Enhanced) is a large-scale bilingual (English and Chinese) dataset for Medical Visual Question Answering (Med-VQA). It consists of radiology images across multiple modalities (CT, MRI, X-Ray) with detailed semantic annotations, comprehensive question-answer pairs, and a structured medical knowledge graph. SLAKE covers five body parts (head, neck, chest, abdomen, and pelvic cavity) and includes both visual reasoning and knowledge-based reasoning tasks. The dataset is specifically designed to advance the development of medical AI systems that can understand clinical images and answer medical questions.
 
-- **Repository:** [More Information Needed]
-- **Paper [optional]:** [More Information Needed]
-- **Demo [optional]:** [More Information Needed]
+- **Curated by:** Bo Liu, Li-Ming Zhan, Li Xu, Lin Ma, Yan Yang, Xiao-Ming Wu from The Hong Kong Polytechnic University, West China Hospital of Sichuan University, and Sichuan Academy of Medical Sciences/Sichuan Provincial People's Hospital
+- **Funded by:** Research grant P0030935 (ZVPY) funded by PolyU (UGC)
+- **Shared by:** The research team from Hong Kong Polytechnic University
+- **Language(s) (NLP):** English (en) 
+- **License:** CC-BY-4.0
 
-## Uses
+### Dataset Sources
 
-<!-- Address questions around how the dataset is intended to be used. -->
+- **Repository:** http://www.med-vqa.com/slake
+- **Dataset repo on Hugging Face:** huggingface.co/datasets/bokelvin/slake
+- **Paper:** [SLAKE: A Semantically-Labeled Knowledge-Enhanced Dataset for Medical Visual Question Answering](https://arxiv.org/abs/2102.09542)
 
-### Direct Use
+## Uses
 
-<!-- This section describes suitable use cases for the dataset. -->
+### Direct Use
 
-[More Information Needed]
+SLAKE is designed for training and evaluating Medical Visual Question Answering (Med-VQA) systems. These systems can be used to:
+- Assist medical professionals in diagnosis by providing a second opinion
+- Help patients better understand their medical imaging results
+- Support clinical education and training of medical students and professionals
+- Serve as a benchmark for measuring progress in medical AI for visual understanding and reasoning
+- Develop multimodal medical AI systems that can integrate visual information with structured medical knowledge
 
 ### Out-of-Scope Use
 
-<!-- This section addresses misuse, malicious use, and uses that the dataset will not work well for. -->
-
-[More Information Needed]
+The dataset should not be used for:
+- Making clinical decisions without expert oversight
+- Directly providing medical advice to patients without physician review
+- Replacing radiologists or other medical professionals
+- Applications outside the medical domain for which the dataset was not designed
+- Claiming diagnostic capabilities without proper clinical validation
 
 ## Dataset Structure
 
-<!-- This section provides a description of the dataset fields, and additional information about the dataset structure such as criteria used to create the splits, relationships between data points, etc. -->
+The SLAKE dataset consists of:
+
+- **Images**: 642 radiology images (282 CT scans, 181 MRI scans, 179 X-rays) covering five body parts (head, neck, chest, abdomen, pelvic cavity)
+
+- **Question-Answer Pairs**: 14,028 QA pairs in both English and Chinese, categorized into 10 content types:
+  - Plane (1,280 questions)
+  - Quality (762 questions)
+  - Modality (1,492 questions)
+  - Position (2,678 questions)
+  - Organ (3,041 questions)
+  - Knowledge Graph (KG) (1,740 questions)
+  - Abnormal (1,696 questions)
+  - Color (647 questions)
+  - Shape (245 questions)
+  - Size (447 questions)
+
+- **Semantic Annotations**:
+  - Segmentation masks for 39 organs and 12 diseases
+  - Bounding box annotations for all important structures
 
-[More Information Needed]
+- **Knowledge Graph**:
+  - 2,603 English and 2,629 Chinese triplets in <head, relation, tail> format
+  - Covers medical concepts like organ functions, disease symptoms, treatments, etc.
+
+- **Dataset Splits**: 70% training (450 images, 9,849 questions), 15% validation (96 images, 2,109 questions), 15% testing (96 images, 2,070 questions)
 
 ## Dataset Creation
 
 ### Curation Rationale
 
-<!-- Motivation for the creation of this dataset. -->
+SLAKE was created to address significant limitations in existing Med-VQA datasets:
+- Lack of semantic labels (segmentation masks, bounding boxes) for training models to find regions of interest
+- Absence of external knowledge resources for answering complex clinical questions
+- Limited coverage of body parts and imaging modalities
+- Need for multilingual support to broaden applications
 
-[More Information Needed]
+The dataset aims to facilitate more sophisticated medical visual reasoning systems that can leverage both visual content and structured medical knowledge.
 
 ### Source Data
 
-<!-- This section describes the source data (e.g. news text and headlines, social media posts, translated sentences, ...). -->
-
 #### Data Collection and Processing
 
-<!-- This section describes the data collection and processing process such as data selection criteria, filtering and normalization methods, tools and libraries used, etc. -->
+- Images were selected from three open-source datasets:
+  1. Medical Segmentation Decathlon
+  2. ChestXray-NIHCC (National Institutes of Health Clinical Center)
+  3. CHAOS (Combined CT-MR Healthy Abdominal Organ Segmentation Challenge)
 
-[More Information Needed]
+- 179 chest X-rays were randomly selected from ChestXray-NIHCC with original disease labels preserved
+- 463 single-slice images were randomly chosen from 3D volume cases in other datasets
+- For the knowledge graph, medical triplets were extracted from OwnThink (a Wikipedia-based knowledge base), then manually filtered and refined
 
 #### Who are the source data producers?
 
-<!-- This section describes the people or systems who originally created the data. It should also include self-reported demographic or identity information for the source data creators if this information is available. -->
-
-[More Information Needed]
+- The original medical images came from public healthcare research repositories
+- The knowledge graph information was derived from OwnThink/Wikipedia
+- Questions and annotations were created by experienced physicians and the research team
 
-### Annotations [optional]
-
-<!-- If the dataset contains annotations which are not part of the initial data collection, use this section to describe them. -->
+### Annotations
 
 #### Annotation process
 
-<!-- This section describes the annotation process such as annotation tools used in the process, the amount of data annotated, annotation guidelines provided to the annotators, interannotator statistics, annotation validation, etc. -->
-
-[More Information Needed]
+- Experienced physicians labeled organs and diseases using ITK-SNAP software
+- The team developed a custom annotation system with question templates for each body part
+- Physicians could choose from candidate questions or amend/rewrite them based on clinical experience
+- Questions were semantically labeled to distinguish between vision-only and knowledge-based types
+- Answer distribution was deliberately balanced to reduce statistical bias (e.g., yes/no questions were designed to have 50-50 distribution)
 
 #### Who are the annotators?
 
-<!-- This section describes the people or systems who created the annotations. -->
-
-[More Information Needed]
+- Experienced physicians performed the image annotations and helped create clinically relevant questions
+- The research team developed the annotation system and constructed the knowledge graph
+- Medical students may have assisted in parts of the annotation process (implied in acknowledgments)
 
 #### Personal and Sensitive Information
 
-<!-- State whether the dataset contains data that might be considered personal, sensitive, or private (e.g., data that reveals addresses, uniquely identifiable names or aliases, racial or ethnic origins, sexual orientations, religious beliefs, political opinions, financial or health data, etc.). If efforts were made to anonymize the data, describe the anonymization process. -->
-
-[More Information Needed]
+The paper states that "This research study was conducted retrospectively using human subject data made available in open access" and that "Ethical approval was not required as confirmed by the license attached with the open access data." The dataset appears to use de-identified medical images from public research repositories.
 
 ## Bias, Risks, and Limitations
 
-<!-- This section is meant to convey both technical and sociotechnical limitations. -->
-
-[More Information Needed]
+- The dataset is limited to five body parts and three imaging modalities
+- Baseline models achieved only ~73% accuracy, indicating significant room for improvement before clinical use
+- The dataset may reflect biases in the original image sources or in the question generation process
+- Despite efforts to balance answers, some inherent biases may remain in the data
+- Performance on this dataset alone should not be used to claim clinical efficacy
+- The knowledge graph, while extensive, cannot capture all relevant medical knowledge
 
 ### Recommendations
 
-<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
-
-Users should be made aware of the risks, biases and limitations of the dataset. More information needed for further recommendations.
+Users should recognize that Med-VQA systems trained on this dataset require clinical validation before deployment in healthcare settings. The dataset should be used primarily for research and development purposes, with careful consideration of potential biases. Evaluation metrics beyond accuracy (such as explainability and clinical relevance) should be considered when developing systems based on this dataset.
 
-## Citation [optional]
-
-<!-- If there is a paper or blog post introducing the dataset, the APA and Bibtex information for that should go in this section. -->
+## Citation
 
 **BibTeX:**
-
-[More Information Needed]
+```
+@article{liu2021slake,
+  title={SLAKE: A Semantically-Labeled Knowledge-Enhanced Dataset for Medical Visual Question Answering},
+  author={Liu, Bo and Zhan, Li-Ming and Xu, Li and Ma, Lin and Yang, Yan and Wu, Xiao-Ming},
+  journal={arXiv preprint arXiv:2102.09542},
+  year={2021}
+}
+```
 
 **APA:**
-
-[More Information Needed]
-
-## Glossary [optional]
-
-<!-- If relevant, include terms and calculations in this section that can help readers understand the dataset or dataset card. -->
-
-[More Information Needed]
-
-## More Information [optional]
-
-[More Information Needed]
-
-## Dataset Card Authors [optional]
-
-[More Information Needed]
-
-## Dataset Card Contact
-
-[More Information Needed]
+Liu, B., Zhan, L.-M., Xu, L., Ma, L., Yang, Y., & Wu, X.-M. (2021). SLAKE: A Semantically-Labeled Knowledge-Enhanced Dataset for Medical Visual Question Answering. arXiv preprint arXiv:2102.09542.