# XXXXXXXXXXXXXXXXX — Data

来源：https://www.kaggle.com/competitions/bone-mineral-density-prediction-challenge/data

本地 Meta Kaggle 快照；不是本次实时页面。

# Dataset Description


This challenge gives you access to a **unique medical dataset** where ordinary **hand and wrist X-rays** become the key to unlocking hidden insights about **hip bone health**. Instead of expensive, immobile DXA machines, these everyday images, combined with a bit of clever AI, could help predict hip **bone mineral density (BMD)** and bring osteoporosis screening closer to everyone, everywhere.

### What’s Inside 

You’ll be working with three splits:

* **train\_groundtruth\_BMD\_only.csv**: 377 training samples with images, patient metadata, and ground-truth hip BMD values.
* **val\_groundtruth\_BMD\_only.csv**: 54 validation samples with the same structure, used to tune and calibrate your models.
* **test\_public\_new\.csv**: 108 unseen test samples with only images and metadata. Ground-truth BMD is hidden and used for leaderboard scoring.

### Data Fields 

Each file includes:

* **image**: File name of the hand or wrist X-ray (PNG format).
* **interview\_age**: Patient age, expressed in **months** (so you can capture fine age differences).
* **Gender**: Biological sex of the patient (M or F).
* **BMD**: Ground-truth hip bone mineral density (training and validation only).

### Example Rows

**Training file (train\_groundtruth\_BMD\_only.csv)**

| image                             | interview\_age (months) | Gender | BMD   |
| --------------------------------- | ----------------------- | ------ | ----- |
| 6.E.1\_9095839\_20090409\_001.png | 732                     | F      | 0.891 |
| 6.C.1\_9100862\_20090219\_001.png | 689                     | M      | 0.877 |

**Test file (test\_public\_new\.csv)**

| image                             | interview\_age (months) | Gender |
| --------------------------------- | ----------------------- | ------ |
| 7.A.1\_9305842\_20100111\_001.png | 745                     | F      |
| 7.B.1\_9401112\_20110322\_001.png | 812                     | M      |