yanzipei / VF-HM

[MICCAI 2023] VF-HM: Vision Loss Estimation using Fundus Photograph for High Myopia

VF-HM

The offical PyTorch implementation of "VF-HM: Vision Loss Estimation using Fundus Photograph for High Myopia" in MICCAI, 2023.

Experiments

Setup

First, prepare your dataset as:

.
└── data.csv         # data file list
    ├── fundus       # fundus dir
    │   ├── aaa.tif  # fundus image
    │   └── bbb.tif  # fundus image
    └── vf           # vf dir
        ├── aaa.json # vf file
        └── bbb.json # vf file

data.csv contains the diagnosis information for each eye, including the fundus file name and corresponding vf file name, eye_type: Left (L) or Right(R); in addition, there is one MM category (C0/C1/C2/C3/C4) for training data only.

fundus_id	vf_id	mm	eye_type
aaa.tif	aaa.json	C0	L
bbb.tif	bbb.json	C1	L
...	...	.	.

fundus dir contains fundus images in tif format with RGB colorful mode. An example of fundus (Left eye) is shown as follows (From Wikipedia):

vf dir contains the vf map in JSON format with 52 effective points. An example of VF (Left eye) is shown as follows:

[[nan nan nan nan nan nan nan nan nan nan]
 [nan nan nan 24. 26. 24. 21. nan nan nan]
 [nan nan 27. 26. 24. 25. 25. 25. nan nan]
 [nan 26. 24. 26. 27. 26. 26. 27. 23. nan]
 [nan 18. nan 25. 29. 29. 29. 28. 26. 23.]
 [nan  0. nan 26. 30. 30. 29. 28. 27. 20.]
 [nan 24. 27. 27. 29. 29. 25. 28. 24. nan]
 [nan nan 27. 29. 29. 29. 28. 29. nan nan]
 [nan nan nan 31. 26. 27. 27. nan nan nan]
 [nan nan nan nan nan nan nan nan nan nan]]

Run

run regression baseline:

bash train_reg.sh > train_reg.out &

run VF-HM:

bash train_vfhm.sh > train_vfhm.out &

Citation

If you use our method or our code in your research, please kindly cite it:

@inproceedings{yan2023vfhm,
title={{VF-HM}: Vision loss estimation using fundus photograph for high myopia},
author={Yan, Zipei and Liang, Dong and Xu, Linchuan and Li, Jiahang and Liu, Zhengji and Wang, Shuai and Cao, Jiannong and Kee, Chea-su},
booktitle={MICCAI},
year={2023},
}