Update README.md

README.md

@@ -1,167 +1,112 @@
+
 ---
 license: mit
 ---
 # 0. FSL_ECG_QA_Dataset
- 
+
 **FSL_ECG_QA_Dataset** is a **benchmark dataset** specifically designed to accompany the paper *"Electrocardiogram–Language Model for Few-Shot Question Answering with Meta Learning"* (**arXiv:2410.14464v1**). It supports research in combining **electrocardiogram (ECG) signals** with **natural language question answering (QA)**, particularly in **few-shot** and **meta-learning** scenarios.
 
 ## 1. Dataset Highlights
- 
+
 - 🧠 **Task Diversification**: Restructured **ECG-QA tasks** promote rapid **few-shot adaptation**.
 - 🧬 **Fusion Mapping**: A lightweight **multimodal mapper** bridges **ECG** and **language features**.
 - 🚀 **Model Generalization**: **LLM-agnostic design** ensures broad **transferability** and **robustness**.
- 
+
 ## 2 Developing Datasets
-# 2.1 Source Datasets
- 
+### 2.1 Source Datasets
+
 The dataset is a structured reorganization of the existing ECG-QA dataset, adapted to suit meta-learning tasks. It draws samples from ECG sources such as [PTB-XL](https://physionet.org/content/ptb-xl/1.0.3/) and [MIMIC-IV-ECG](https://physionet.org/content/mimic-iv-ecg/1.0/), and [ECG-QA dataset](https://github.com/Jwoo5/ecg-qa?tab=readme-ov-file) organizes them into diverse task sets based on question types including **verify(yes/no)**, **choice(Condition_A/Condition_B)**, and **query(open-ended)** question in table 2.2. and clinical attributes (e.g., SCP codes, noise type, axis deviation) used to describing the ECG. This structure enables models to rapidly adapt to new diagnostic tasks with limited annotated examples.
 
 <img src="img/distribution_attr.png" alt="Figure 1: Illustration of class formation and attribute distribution for different question types." width="600"/>
 
-# 2.2 Dataset Description
-Overview of question types and data distribution within the meta learning benchmark dataset created for few-shot ECG question answering.
+### 2.2 Dataset Description
 
-### Supported Tasks
+Overview of question types and data distribution within the meta learning benchmark dataset created for few-shot ECG question answering.
 
--- [CUSTOMIZE THIS] Describe the tasks this dataset supports:
-- What tasks can be performed on this dataset? 
+#### Supported Tasks
 
+- What tasks can be performed on this dataset?
 - Are there any code associated with this dataset?
 - Electrocardiogram–Language Model for Few-Shot Question Answering with Meta Learning(ICASSP 2025), https://arxiv.org/html/2410.14464v1
 
-
 **class(template_id_attribute_answer pairs)**
 
-### Data Instances
+#### Data Instances
+
+Merge all data in train/val/test dataset in ECG-QA and change it into different class, Use a consistent ID system to track attribute-answer combinations (definition of "way" in meta-learning):
 
-<!-- [CUSTOMIZE THIS] Provide examples of data instances from the dataset. Include:
-- A description of what each instance represents
-- One or more concrete examples in JSON or dictionary format -->
-Merge all data in train/val/test dataset in ECG-QA and change it into different class, Use a consistent ID system to track attribute-answer combinations(defination of "way" in meta-learning):
-example in ecg-qa:
 ```python
-# Example data instance
-# Example data instance
 {
-    "template_id": 1,  # ID of the template used
-    "question_id": 0,  # Unique identifier for the question
-    "sample_id": 0,    # Unique identifier for the data sample
-    "question_type": "single-verify",  # Type of question
-    "attribute_type": "scp_code",  # Type of attribute being verified
-    "question": "Is there evidence of non-diagnostic t abnormalities on this ECG?",  # Natural language question
-    "answer": [
-        "yes"  # Ground truth answer
-    ],
-    "ecg_id": [
-        12662  # ID referencing the ECG record
-    ],
-    "attribute": [
-        "non-diagnostic t abnormalities"  # The specific attribute being evaluated
-    ]
+    "template_id": 1,
+    "question_id": 0,
+    "sample_id": 0,
+    "question_type": "single-verify",
+    "attribute_type": "scp_code",
+    "question": "Is there evidence of non-diagnostic t abnormalities on this ECG?",
+    "answer": ["yes"],
+    "ecg_id": [12662],
+    "attribute": ["non-diagnostic t abnormalities"]
 }
-
 ```
 
-## new class
-Example: 5_atrial_fibrillation_yes represents Template ID *5*, attribute *"atrial fibrillation"*, answer *"yes"*.
-
-## class build
-
-
-The output and details amount of dataset in different question type.
-<div style="width: 80%; font-size: 14px;">
-| Question Type | Attributes | Answers | Classes (train:test) | Samples | Example |
-|---------------|------------|---------|---------------------|---------|---------|
-| Single-Verify | 94 | yes/no | 156 (124:32) | 34,105 | Q: Does this ECG show 1st degree av block? <br> A: yes/no |
-| Single-Choose | 165 | both/none/attr_1/attr_2 | 262 (209:53) | 47,655 | Q: Which noise does this ECG show, baseline drift or static noise? <br> A: baseline drift/static noise/both/none |
-| Single-Query | 30 | attr_1/attr_2/.../attr_n | 260 (208:52) | 63,125 | Q: What direction is this ECG deviated to? <br> A: Normal axis/.../open-ending |
-| All | 206 | yes/no/both/none/.../attr_n | 678 (541:137) | 144,885 | ... |
-</div>
-
-### Loading the Dataset
-
-```python
-
-# Example code to load the dataset ptb-xl
+#### New Class Naming
 
-python load_class.py \
-  --base_path /your/actual/path/to/ecgqa/ptbxl/paraphrased \
-  --test_dataset ptb-xl
+Example: `5_atrial_fibrillation_yes` represents Template ID *5*, attribute *"atrial fibrillation"*, answer *"yes"*.
 
-# Example code to load the dataset mimiv-iv-ecg
+#### Class Build
 
-# the random selected 30k ecg from mimic-iv-ecg used to test
-processed_test_30k.json
+| Question Type   | Attributes | Answers                          | Classes (train:test) | Samples | Example                                                                 |
+|-----------------|------------|----------------------------------|-----------------------|---------|-------------------------------------------------------------------------|
+| Single-Verify   | 94         | yes/no                           | 156 (124:32)          | 34,105  | Q: Does this ECG show 1st degree av block? <br> A: yes/no               |
+| Single-Choose   | 165        | both/none/attr_1/attr_2          | 262 (209:53)          | 47,655  | Q: Which noise does this ECG show, baseline drift or static noise?     |
+| Single-Query    | 30         | attr_1/attr_2/.../attr_n         | 260 (208:52)          | 63,125  | Q: What direction is this ECG deviated to? <br> A: Normal axis/...     |
+| All             | 206        | yes/no/both/none/.../attr_n      | 678 (541:137)         | 144,885 | ...                                                                     |
 
-python load_class.py \
-  --paraphrased_path /your/actual/path/to/ecgqa/mimic-iv-ecg/paraphrased \
-  --test_dataset mimic
+#### Loading the Dataset
 
+```bash
+python load_class.py --base_path /your/actual/path/to/ecgqa/ptbxl/paraphrased --test_dataset ptb-xl
+python load_class.py --paraphrased_path /your/actual/path/to/ecgqa/mimic-iv-ecg/paraphrased --test_dataset mimic
 ```
 
-# Access splits
+```python
 train_data = dataset["train"]
 validation_data = dataset["validation"]
 test_data = dataset["test"]
 
-# Example usage
 for example in train_data.select(range(3)):
     print(example)
 ```
-```bash
-
-
-
-
-## few-shot build
 
+### Few-shot Build
 
 ```bash
-python data_loader.py \
-  --paraphrased_path /your/actual/path/to/ecgqa/mimic-iv-ecg/paraphrased \
-  --test_dataset mimic
+python data_loader.py --paraphrased_path /your/actual/path/to/ecgqa/mimic-iv-ecg/paraphrased --test_dataset mimic
 ```
 
-<img src="img/FSL_ECG_QAMeta-Learning.png" alt="An example of n way 1 shot and how to utilize in Meta-learning." style="max-width:100%; height:auto;">
+<img src="img/FSL_ECG_QAMeta-Learning.png" alt="Few-shot Meta-learning Example" width="600"/>
 
-In summary, FSL_ECG_QA_Dataset serves as a powerful benchmark for developing robust and generalizable ECG-based QA systems in data-scarce clinical environments. 
 
 ## Dataset Structure
 
 ### Data Fields
 
-<!-- [CUSTOMIZE THIS] Describe all data fields, including:
-- Field name
-- Data type
-- Description of what the field represents
-- For categorical fields, the possible values and their meanings -->
-
 - `feature1`: a `string` feature representing <!-- description -->
 - `feature2`: a `string` feature representing <!-- description -->
 - `label`: a `int64` classification label, with 0 indicating <!-- meaning --> and 1 indicating <!-- meaning -->
 
 ### Data Splits
 
--- Describe how the data is split:
 - Number of instances in each split (train/test): 8:2
-- Criteria used for splitting the data
-- first split based on template id, this will keep there's no overlap between expression between training and testing set.
-- then split for support and query set: since there's aboudant amout of data randomly select sample from same question class for support and query set.
+- Criteria: first split based on template id (no expression overlap between train/test), then random split for support/query set in few-shot tasks.
 
 ## Dataset Creation
 
 ### Curation Rationale
 
-- What need does it address?
-Developing robust and reliable multimodal QA systems for ECG interpretation relies on the availability of both high-quality and large quantities of labeled data. Yet, obtaining massive amounts of labeled ECGs from cardiologists is costly, which often results in limited datasets. Traditional supervised learning methods tend to perform well only on data with the same distribution as the training data. In real-world deployment, however, models frequently encounter new tasks and previously unseen populations outside the training distribution, where traditional methods may fail. Meta-learning, a paradigm focused on “learning to learn”, offers a compelling solution to this challenge. By training models on a diverse range of tasks, meta-learning enables them to acquire transferable knowledge and adapt rapidly to new, unseen tasks with minimal labeled data. This adaptive capacity is particularly valuable in the ECG-language QA domain, where new diagnostic questions and data distributions constantly emerge.
-
-
+Developing robust and reliable multimodal QA systems for ECG interpretation relies on the availability of both high-quality and large quantities of labeled data. Meta-learning, a paradigm focused on “learning to learn”, enables them to acquire transferable knowledge and adapt rapidly to new, unseen tasks with minimal labeled data.
 
-### Citation Information
-
-<!-- [CUSTOMIZE THIS] Provide citation information:
-- How should this dataset be cited?
-- BibTeX citation -->
+## Citation
 
 ```
 @inproceedings{10888594,
@@ -169,23 +114,18 @@ Developing robust and reliable multimodal QA systems for ECG interpretation reli
   booktitle={ICASSP 2025 - 2025 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)}, 
   title={Electrocardiogram Report Generation and Question Answering via Retrieval-Augmented Self-Supervised Modeling}, 
   year={2025},
-  volume={},
-  number={},
   pages={1-5},
-  keywords={Electric potential;Accuracy;Large language models;MIMICs;Self-supervised learning;Electrocardiography;Signal processing;Question answering (information retrieval);Speech processing;Investment;electrocardiogram;retrieval augmented generation;self-supervised learning;large language models},
-  doi={10.1109/ICASSP49660.2025.10888594}}
+  doi={10.1109/ICASSP49660.2025.10888594}
+}
 ```
 
 ## How to Use
 
-
 ### Example Preprocessing and Training
 
 ```python
-# Example preprocessing and model training code
 from transformers import AutoTokenizer, AutoModelForSequenceClassification, Trainer, TrainingArguments
 
-# Load tokenizer and tokenize data
 tokenizer = AutoTokenizer.from_pretrained("bert-base-uncased")
 
 def tokenize_function(examples):
@@ -193,10 +133,8 @@ def tokenize_function(examples):
 
 tokenized_dataset = dataset.map(tokenize_function, batched=True)
 
-# Define model
 model = AutoModelForSequenceClassification.from_pretrained("bert-base-uncased", num_labels=2)
 
-# Define training arguments
 training_args = TrainingArguments(
     output_dir="./results",
     per_device_train_batch_size=16,
@@ -207,7 +145,6 @@ training_args = TrainingArguments(
     load_best_model_at_end=True,
 )
 
-# Define trainer
 trainer = Trainer(
     model=model,
     args=training_args,
@@ -215,7 +152,5 @@ trainer = Trainer(
     eval_dataset=tokenized_dataset["validation"],
 )
 
-# Train model
 trainer.train()
 ```
-