README.md

---
pretty_name: Lower-limb Alignment Measurements for MRKR Subset
language:
  - en
license: cc-by-nc-4.0
task_categories:
  - tabular-regression
size_categories:
  - n<1K
tags:
  - medical
  - radiograph
  - knee
  - osteoarthritis
  - lower-limb-alignment
  - intra-rater-reliability
---

# Lower-limb Alignment Measurements for the MRKR Subset

Anonymised knee radiograph metadata with manual and derived radiographic alignment
measurements for a subset of the Emory Knee Radiograph (MRKR) dataset [1]. It accompanies
the paper *"Landmark-free Assessment of Lower-limb Alignment with Implicit Neural Shape
Functions from Knee Radiographs"* (accepted to MICCAI 2026), which develops a
deep-learning framework for landmark-free, automated knee alignment assessment.

![Framework overview](./pics/INSF_overview.png)

Release file: `data/metadata_with_measurements_final.csv` — **402 rows**, **14 columns**,
**244 unique anonymised patient IDs**.

## Dataset Details

<table>
<tr>
<td valign="top">

The dataset provides three angular alignment measurements:

- **aTFA** — anatomical tibiofemoral angle, formed by the anatomical axes of the femur and tibia.
- **aMPTA** — anatomical medial proximal tibial angle, formed by the anatomical tibial axis and the tibial joint line (medial side).
- **JLCA** — joint line convergence angle, formed by the femoral and tibial joint lines.

A single board-certified radiologist recorded **absolute aTFA**, **acute aMPTA**, and **absolute JLCA** over two reading rounds 7–10 days apart, with the second reading blinded to the first; per-round values (`_r1`, `_r2`) and their mean (`_mean`) are provided. Only the de-identified measurement/metadata table is released, not the images.

</td>
<td valign="top" width="320">
<img src="./pics/Knee_angles.jpg" alt="Knee angles (aTFA, aMPTA, JLCA)" width="300">
</td>
</tr>
</table>

- **Curated by:** Zhisen Hu, Antti Kemppainen, David Johnson, Egor Panfilov, Huy Hoang Nguyen, Timothy Cootes, Claudia Lindner, Aleksei Tiulpin
- **Language(s):** English (en) — column labels and textual metadata
- **Modality / format:** Tabular knee radiograph metadata and measurements (CSV)
- **License:** CC BY-NC 4.0
- **Paper:** Hu et al., arXiv:2606.15250 (accepted to MICCAI 2026, poster)
- **Source images:** Emory Knee Radiograph (MRKR) dataset [1]
- **Repository / Contact:** [More Information Needed]

## Uses

**Direct use** (non-commercial research and education): alignment-measurement research
(aTFA, aMPTA, JLCA), osteoarthritis imaging-metadata analysis, model development and
evaluation, and intra-rater reliability studies using the `_r1`/`_r2`/`_mean` columns.

**Out-of-scope:** clinical decision-making, diagnosis, or treatment; commercial use
without prior written permission; and population-level or epidemiological inference (this
is a randomly drawn subset).

## Dataset Structure

Single CSV (402 × 14); column-level definitions in `docs/DATA_DICTIONARY.md`. Key fields:

- `empi_anon` — anonymised patient identifier. Bilateral studies were split into separate
  left and right knees, so a patient may contribute two rows; this is why 402 knees map to
  244 unique IDs.
- `laterality` — knee side (left / right).
- `aTFA_r1`, `aTFA_r2`, `aTFA_mean` — absolute anatomical tibiofemoral angle.
- `aMPTA_r1`, `aMPTA_r2`, `aMPTA_mean` — acute anatomical medial proximal tibial angle.
- `JLCA_r1`, `JLCA_r2`, `JLCA_mean` — absolute joint line convergence angle.

Remaining columns are DICOM / image-derived metadata (see the data dictionary).
**Splits:** none defined. **Missing values:** 15 rows are missing several DICOM / image
metadata fields.

## Dataset Creation

The 402 knees were randomly selected from the MRKR dataset [1], comprising both bilateral
and unilateral studies. Bilateral studies were split into individual knees. A single
board-certified radiologist measured aTFA, aMPTA, and JLCA on two occasions **7–10 days
apart**, with the second reading blinded to the first, to enable intra-rater consistency
assessment; `_r1` and `_r2` hold the two readings and `_mean` their average. The data is
de-identified (only `empi_anon` plus DICOM/image metadata) and was released with approval
from the relevant data owner, supervisor, or institution.

## Bias, Risks, and Limitations

Measurements come from a single rater (supporting intra- but not inter-rater analysis) and
are subject to manual variability. The accompanying paper reports the MRKR subset as
roughly 40% White and 40% African American patients, so it represents a specific cohort and
may not generalise to other populations; 15 rows also have missing metadata. Use for
non-commercial research and education only; do not use for clinical decision-making.

## Citation

You must cite the following paper (accepted to MICCAI 2026):

```bibtex
@inproceedings{hu2026landmark,
  title={Landmark-free Assessment of Lower-limb Alignment with Implicit Neural Shape Functions from Knee Radiographs},
  author={Hu, Zhisen and Kemppainen, Antti and Johnson, David and Panfilov, Egor and Nguyen, Huy Hoang and Cootes, Timothy and Lindner, Claudia and Tiulpin, Aleksei},
  booktitle={International Conference on Medical Image Computing and Computer-Assisted Intervention (MICCAI)},
  year={2026},
  organization={Springer}
}
```

**APA:** Hu, Z., Kemppainen, A., Johnson, D., Panfilov, E., Nguyen, H. H., Cootes, T.,
Lindner, C., & Tiulpin, A. (2026). *Landmark-free assessment of lower-limb alignment with
implicit neural shape functions from knee radiographs.* In *Medical Image Computing and
Computer-Assisted Intervention (MICCAI)*. Springer.

## Glossary

aTFA — anatomical tibiofemoral angle; aMPTA — anatomical medial proximal tibial angle;
JLCA — joint line convergence angle; `_r1`/`_r2` — first/second reading rounds (7–10 days
apart); `_mean` — mean of the two; `empi_anon` — anonymised patient identifier.

## Reference

[1] Price BJ, Gichoya J, Chavoshi M, Adleberg J, Thomas K, Zaiman Z, Mansuri A,
Brown-Mulry B, Okechukwu C, Dapamede T, Li F, Isaac RS, Goltzer O, Gottschalk M,
Trivedi H. *The Emory Knee Radiograph (MRKR) Dataset.* J Imaging Inform Med. 2026 Mar 10.
doi: 10.1007/s10278-026-01902-6. Epub ahead of print. PMID: 41807882.