File size: 24,172 Bytes
873e8c1 b54fd81 873e8c1 b54fd81 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 |
---
license: mit
language:
- en
datasets:
- mathpluscode/ACDC
tags:
- medical
- cardiac
- MRI
- foundation model
- MAE
---
# CineMA - A Foundation Model for Cine Cardiac Magnetic Resonance Images 🎥🫀
This repository contains the weights for **CineMA**, a foundation model for **Cine** cardiac magnetic resonance (CMR)
imaging based on **M**asked-**A**utoencoder. The model was pre-trained on over 74,000 pairs of short-axis and long-axis
cine CMR images from the UK Biobank.
CineMA was evaluated across a diverse range of clinically relevant downstream tasks, including
- Ventricle and myocardium segmentation
- Cardiovascular disease (CVD) detection and classification
- Patient sex classification
- CMR machine vendor classification
- Ejection fraction (EF) regression
- Patient body mass index (BMI) regression
- Patient age regression
- Mid-ventricular and apical landmark localization
These tasks were studied across multiple datasets:
- [ACDC](https://www.creatis.insa-lyon.fr/Challenge/acdc/)
- [M&Ms](https://www.ub.edu/mnms/)
- [M&Ms2](https://www.ub.edu/mnms-2/)
- [Kaggle](https://www.kaggle.com/c/second-annual-data-science-bowl/data)
- [Rescan](https://www.ahajournals.org/doi/full/10.1161/CIRCIMAGING.119.009214)
- [Landmark](https://pubs.rsna.org/doi/10.1148/ryai.2021200197)
Compared to convolutional neural network baselines such as UNet and ResNet, CineMA demonstrated superior or comparable
performance, especially in sample efficiency and generalization to out-of-distribution data not seen during pretraining
or fine-tuning.
By releasing the model weights and code for pretraining, fine-tuning, and inference, CineMA aims to lower the barrier to
entry for cardiac imaging research, foster reproducibility, and encourage broader adoption across institutions.
➡️ **Manuscript:** [TBD](https://arxiv.org/)
➡️ **Code:** [mathpluscode/CineMA](https://github.com/mathpluscode/CineMA)
## Fine-tuned CineMA Models
The filenames of fine-tuned model weights follow the convention of `finetuned/<task>/<data>_<view>_<seed>.safetensors`
where number 0, 1, and 2 correspond to the different training seeds.
Check the "Inference Example" column to see example inference scripts using these trained models.
| Training Task | Training Data | Input View | Input Timeframes | Model Weights and Configurations | Inference Example |
| ----------------------------------------------- | ------------- | ---------- | ---------------- | -------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------------------------------------------------------------------ |
| Ventricle and myocardium segmentation | ACDC | SAX | 1 | [finetuned/segmentation/acdc_sax_0.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/segmentation/acdc_sax_0.safetensors)<br>[finetuned/segmentation/acdc_sax_1.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/segmentation/acdc_sax_1.safetensors)<br>[finetuned/segmentation/acdc_sax_2.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/segmentation/acdc_sax_2.safetensors)<br>[finetuned/segmentation/sax.yaml](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/segmentation/sax.yaml) | [segmentation_sax.py](https://github.com/mathpluscode/CineMA/blob/main/examples/inference/segmentation_sax.py) |
| Ventricle and myocardium segmentation | M&Ms | SAX | 1 | [finetuned/segmentation/mnms_sax_0.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/segmentation/mnms_sax_0.safetensors)<br>[finetuned/segmentation/mnms_sax_1.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/segmentation/mnms_sax_1.safetensors)<br>[finetuned/segmentation/mnms_sax_2.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/segmentation/mnms_sax_2.safetensors)<br>[finetuned/segmentation/sax.yaml](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/segmentation/sax.yaml) | [segmentation_sax.py](https://github.com/mathpluscode/CineMA/blob/main/examples/inference/segmentation_sax.py) |
| Ventricle and myocardium segmentation | M&Ms2 | SAX | 1 | [finetuned/segmentation/mnms2_sax_0.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/segmentation/mnms2_sax_0.safetensors)<br>[finetuned/segmentation/mnms2_sax_1.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/segmentation/mnms2_sax_1.safetensors)<br>[finetuned/segmentation/mnms2_sax_2.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/segmentation/mnms2_sax_2.safetensors)<br>[finetuned/segmentation/sax.yaml](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/segmentation/sax.yaml) | [segmentation_sax.py](https://github.com/mathpluscode/CineMA/blob/main/examples/inference/segmentation_sax.py) |
| Ventricle and myocardium segmentation | M&Ms2 | LAX 4C | 1 | [finetuned/segmentation/mnms2_lax_4c_0.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/segmentation/mnms2_lax_4c_0.safetensors)<br>[finetuned/segmentation/mnms2_lax_4c_1.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/segmentation/mnms2_lax_4c_1.safetensors)<br>[finetuned/segmentation/mnms2_lax_4c_2.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/segmentation/mnms2_lax_4c_2.safetensors)<br>[finetuned/segmentation/lax_4c.yaml](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/segmentation/lax_4c.yaml) | [segmentation_lax_4c.py](https://github.com/mathpluscode/CineMA/blob/main/examples/inference/segmentation_lax_4c.py) |
| CVD classification | ACDC | SAX | 2 (ED and ES) | [finetuned/classification_cvd/acdc_sax_0.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/classification_cvd/acdc_sax_0.safetensors)<br>[finetuned/classification_cvd/acdc_sax_1.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/classification_cvd/acdc_sax_1.safetensors)<br>[finetuned/classification_cvd/acdc_sax_2.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/classification_cvd/acdc_sax_2.safetensors)<br>[finetuned/classification_cvd/acdc_sax.yaml](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/classification_cvd/acdc_sax.yaml) | [classification_cvd.py](https://github.com/mathpluscode/CineMA/blob/main/examples/inference/classification_cvd.py) |
| CVD classification | M&Ms | SAX | 2 (ED and ES) | [finetuned/classification_cvd/mnms_sax_0.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/classification_cvd/mnms_sax_0.safetensors)<br>[finetuned/classification_cvd/mnms_sax_1.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/classification_cvd/mnms_sax_1.safetensors)<br>[finetuned/classification_cvd/mnms_sax_2.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/classification_cvd/mnms_sax_2.safetensors)<br>[finetuned/classification_cvd/mnms_sax.yaml](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/classification_cvd/mnms_sax.yaml) | [classification_cvd.py](https://github.com/mathpluscode/CineMA/blob/main/examples/inference/classification_cvd.py) |
| CVD classification | M&Ms2 | SAX | 2 (ED and ES) | [finetuned/classification_cvd/mnms2_sax_0.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/classification_cvd/mnms2_sax_0.safetensors)<br>[finetuned/classification_cvd/mnms2_sax_1.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/classification_cvd/mnms2_sax_1.safetensors)<br>[finetuned/classification_cvd/mnms2_sax_2.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/classification_cvd/mnms2_sax_2.safetensors)<br>[finetuned/classification_cvd/mnms2_sax.yaml](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/classification_cvd/mnms2_sax.yaml) | [classification_cvd.py](https://github.com/mathpluscode/CineMA/blob/main/examples/inference/classification_cvd.py) |
| CVD classification | M&Ms2 | LAX 4C | 2 (ED and ES) | [finetuned/classification_cvd/mnms2_lax_4c_0.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/classification_cvd/mnms2_lax_4c_0.safetensors)<br>[finetuned/classification_cvd/mnms2_lax_4c_1.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/classification_cvd/mnms2_lax_4c_1.safetensors)<br>[finetuned/classification_cvd/mnms2_lax_4c_2.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/classification_cvd/mnms2_lax_4c_2.safetensors)<br>[finetuned/classification_cvd/mnms2_lax_4c.yaml](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/classification_cvd/mnms2_lax_4c.yaml) | [classification_cvd.py](https://github.com/mathpluscode/CineMA/blob/main/examples/inference/classification_cvd.py) |
| Patient sex classification | M&Ms | SAX | 2 (ED and ES) | [finetuned/classification_sex/mnms_sax_0.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/classification_sex/mnms_sax_0.safetensors)<br>[finetuned/classification_sex/mnms_sax_1.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/classification_sex/mnms_sax_1.safetensors)<br>[finetuned/classification_sex/mnms_sax_2.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/classification_sex/mnms_sax_2.safetensors)<br>[finetuned/classification_sex/mnms_sax.yaml](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/classification_sex/mnms_sax.yaml) | [classification_sex.py](https://github.com/mathpluscode/CineMA/blob/main/examples/inference/classification_sex.py) |
| CMR machine vendor classification | M&Ms2 | SAX | 2 (ED and ES) | [finetuned/classification_vendor/mnms2_sax_0.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/classification_vendor/mnms2_sax_0.safetensors)<br>[finetuned/classification_vendor/mnms2_sax_1.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/classification_vendor/mnms2_sax_1.safetensors)<br>[finetuned/classification_vendor/mnms2_sax_2.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/classification_vendor/mnms2_sax_2.safetensors)<br>[finetuned/classification_vendor/mnms2_sax.yaml](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/classification_vendor/mnms2_sax.yaml) | [classification_vendor.py](https://github.com/mathpluscode/CineMA/blob/main/examples/inference/classification_vendor.py) |
| CMR machine vendor classification | M&Ms2 | LAX 4C | 2 (ED and ES) | [finetuned/classification_vendor/mnms2_lax_4c_0.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/classification_vendor/mnms2_lax_4c_0.safetensors)<br>[finetuned/classification_vendor/mnms2_lax_4c_1.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/classification_vendor/mnms2_lax_4c_1.safetensors)<br>[finetuned/classification_vendor/mnms2_lax_4c_2.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/classification_vendor/mnms2_lax_4c_2.safetensors)<br>[finetuned/classification_vendor/mnms2_lax_4c.yaml](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/classification_vendor/mnms2_lax_4c.yaml) | [classification_vendor.py](https://github.com/mathpluscode/CineMA/blob/main/examples/inference/classification_vendor.py) |
| EF regression | ACDC | SAX | 2 (ED and ES) | [finetuned/regression_ef/acdc_sax_0.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/regression_ef/acdc_sax_0.safetensors)<br>[finetuned/regression_ef/acdc_sax_1.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/regression_ef/acdc_sax_1.safetensors)<br>[finetuned/regression_ef/acdc_sax_2.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/regression_ef/acdc_sax_2.safetensors)<br>[finetuned/regression_ef/acdc_sax.yaml](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/regression_ef/acdc_sax.yaml) | [regression_ef.py](https://github.com/mathpluscode/CineMA/blob/main/examples/inference/regression_ef.py) |
| EF regression | M&Ms | SAX | 2 (ED and ES) | [finetuned/regression_ef/mnms_sax_0.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/regression_ef/mnms_sax_0.safetensors)<br>[finetuned/regression_ef/mnms_sax_1.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/regression_ef/mnms_sax_1.safetensors)<br>[finetuned/regression_ef/mnms_sax_2.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/regression_ef/mnms_sax_2.safetensors)<br>[finetuned/regression_ef/mnms_sax.yaml](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/regression_ef/mnms_sax.yaml) | [regression_ef.py](https://github.com/mathpluscode/CineMA/blob/main/examples/inference/regression_ef.py) |
| EF regression | M&Ms2 | SAX | 2 (ED and ES) | [finetuned/regression_ef/mnms2_sax_0.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/regression_ef/mnms2_sax_0.safetensors)<br>[finetuned/regression_ef/mnms2_sax_1.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/regression_ef/mnms2_sax_1.safetensors)<br>[finetuned/regression_ef/mnms2_sax_2.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/regression_ef/mnms2_sax_2.safetensors)<br>[finetuned/regression_ef/mnms2_sax.yaml](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/regression_ef/mnms2_sax.yaml) | [regression_ef.py](https://github.com/mathpluscode/CineMA/blob/main/examples/inference/regression_ef.py) |
| EF regression | M&Ms2 | LAX 4C | 2 (ED and ES) | [finetuned/regression_ef/mnms2_lax_4c_0.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/regression_ef/mnms2_lax_4c_0.safetensors)<br>[finetuned/regression_ef/mnms2_lax_4c_1.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/regression_ef/mnms2_lax_4c_1.safetensors)<br>[finetuned/regression_ef/mnms2_lax_4c_2.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/regression_ef/mnms2_lax_4c_2.safetensors)<br>[finetuned/regression_ef/mnms2_lax_4c.yaml](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/regression_ef/mnms2_lax_4c.yaml) | [regression_ef.py](https://github.com/mathpluscode/CineMA/blob/main/examples/inference/regression_ef.py) |
| Patient BMI regression | ACDC | SAX | 2 (ED and ES) | [finetuned/regression_bmi/acdc_sax_0.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/regression_bmi/acdc_sax_0.safetensors)<br>[finetuned/regression_bmi/acdc_sax_1.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/regression_bmi/acdc_sax_1.safetensors)<br>[finetuned/regression_bmi/acdc_sax_2.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/regression_bmi/acdc_sax_2.safetensors)<br>[finetuned/regression_bmi/acdc_sax.yaml](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/regression_bmi/acdc_sax.yaml) | [regression_bmi.py](https://github.com/mathpluscode/CineMA/blob/main/examples/inference/regression_bmi.py) |
| Patient age regression | M&Ms | SAX | 2 (ED and ES) | [finetuned/regression_age/mnms_sax_0.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/regression_age/mnms_sax_0.safetensors)<br>[finetuned/regression_age/mnms_sax_1.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/regression_age/mnms_sax_1.safetensors)<br>[finetuned/regression_age/mnms_sax_2.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/regression_age/mnms_sax_2.safetensors)<br>[finetuned/regression_age/mnms_sax.yaml](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/regression_age/mnms_sax.yaml) | [regression_age.py](https://github.com/mathpluscode/CineMA/blob/main/examples/inference/regression_age.py) |
| Landmark localization by heatmap regression | Landmark | LAX 2C | 1 | [finetuned/landmark_heatmap/lax_2c_0.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/landmark_heatmap/lax_2c_0.safetensors)<br>[finetuned/landmark_heatmap/lax_2c_1.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/landmark_heatmap/lax_2c_1.safetensors)<br>[finetuned/landmark_heatmap/lax_2c_2.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/landmark_heatmap/lax_2c_2.safetensors)<br>[finetuned/landmark_heatmap/lax_2c.yaml](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/landmark_heatmap/lax_2c.yaml) | [landmark_heatmap.py](https://github.com/mathpluscode/CineMA/blob/main/examples/inference/landmark_heatmap.py) |
| Landmark localization by heatmap regression | Landmark | LAX 4C | 1 | [finetuned/landmark_heatmap/lax_4c_0.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/landmark_heatmap/lax_4c_0.safetensors)<br>[finetuned/landmark_heatmap/lax_4c_1.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/landmark_heatmap/lax_4c_1.safetensors)<br>[finetuned/landmark_heatmap/lax_4c_2.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/landmark_heatmap/lax_4c_2.safetensors)<br>[finetuned/landmark_heatmap/lax_4c.yaml](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/landmark_heatmap/lax_4c.yaml) | [landmark_heatmap.py](https://github.com/mathpluscode/CineMA/blob/main/examples/inference/landmark_heatmap.py) |
| Landmark localization by coordinates regression | Landmark | LAX 2C | 1 | [finetuned/landmark_coordinate/lax_2c_0.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/landmark_coordinate/lax_2c_0.safetensors)<br>[finetuned/landmark_coordinate/lax_2c_1.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/landmark_coordinate/lax_2c_1.safetensors)<br>[finetuned/landmark_coordinate/lax_2c_2.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/landmark_coordinate/lax_2c_2.safetensors)<br>[finetuned/landmark_coordinate/lax_2c.yaml](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/landmark_coordinate/lax_2c.yaml) | [landmark_coordinate.py](https://github.com/mathpluscode/CineMA/blob/main/examples/inference/landmark_coordinate.py) |
| Landmark localization by coordinates regression | Landmark | LAX 4C | 1 | [finetuned/landmark_coordinate/lax_4c_0.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/landmark_coordinate/lax_4c_0.safetensors)<br>[finetuned/landmark_coordinate/lax_4c_1.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/landmark_coordinate/lax_4c_1.safetensors)<br>[finetuned/landmark_coordinate/lax_4c_2.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/landmark_coordinate/lax_4c_2.safetensors)<br>[finetuned/landmark_coordinate/lax_4c.yaml](https://huggingface.co/mathpluscode/CineMA/blob/main/finetuned/landmark_coordinate/lax_4c.yaml) | [landmark_coordinate.py](https://github.com/mathpluscode/CineMA/blob/main/examples/inference/landmark_coordinate.py) |
## Pre-trained CineMA Model
The pre-trained CineMA model backbone is available at
[pretrained/cinema.safetensors](https://huggingface.co/mathpluscode/CineMA/blob/main/pretrained/cinema.safetensors) with
configuration [pretrained/cinema.yaml](https://huggingface.co/mathpluscode/CineMA/blob/main/pretrained/cinema.yaml).
Following scripts demonstrated how to fine-tune this backbone using
[a preprocessed version of ACDC dataset](https://huggingface.co/datasets/mathpluscode/ACDC):
- [Ventricle and myocardium segmentation](https://github.com/mathpluscode/CineMA/blob/main/examples/train/segmentation.py)
- [Cardiovascular disease classification](https://github.com/mathpluscode/CineMA/blob/main/examples/train/classification.py)
- [Ejection fraction regression](https://github.com/mathpluscode/CineMA/blob/main/examples/train/regression.py)
## Citation
## Contact
For questions or collaborations, please contact Yunguan Fu (yunguan.fu.18@ucl.ac.uk). |