Prediction of visual acuity improvement after macular hole surgery

The data represent successful macular hole surgeries from 2014 to 2018 at CHU de Québec. In total, the dataset contains 2658 OCT scans (images) from 493 patients along with their clinical data. It is difficult for a clinician to predict what the long-term impact of this operation will be on a patient's visual acuity. This is where the operation is interesting to analyze from a machine learning perspective. We are interested in knowing if we can, from examples of surgery, predict to what extent the operation will have a positive impact on the patient's visual acuity.

For each patient, we have pre-operative clinical data. Also, at different pre- and post-operative times, we have a visual acuity measurement as well as two images (horizontal and vertical OCT scans) of the eye with a macular hole.

Brief description of the data set

The data set is separated to match the reference article from Godbout et al., (2021). It consists of 4 folders:

• train: Training set in the article
• validation: Validation set in the article
• test: Test set in the article
• others: Data set where the measured vision at 6 months is not available (not in the article)

For each of the folders, a clinical_data.csv file contains information about each patient in the record. The csvfile has some missing data, which can be labelled as -9, -7 or simply be absent.

An octs file is also present. It contains all the OCTs owned for the patients in the dataset, stored in the format {patient_id}_{time}_{orientation}.tiff, where orientation is either "H" or "V", respectively for horizontal and vertical scans.

IMPORTANT: Some OCTs are missing. Most of the time, horizontal and vertical OCTs of a patient are available for each timestep for which a visual acuity measurement is available, but this is not always the case. Therefore, the possible absence of OCTs must be taken into account when processing the dataset.

Features description

• id: Unique identifier of the patient.
• age: Age of the patient.
• pseudophakic: Characteristic describing if a patient has received a cataract surgery in the past. 1 if yes, 0 otherwise.
• mh_duration: Duration (in weeks) of the patient's macular hole.
• elevated_edge: Presence or not of elevated edge in the patient. 1 if yes, 0 otherwise.
• mh_size: Size (in micrometers) of the macular hole before the operation.
• VA_baseline: Visual acuity (measured in ETDRS letters) of the patient before surgery.
• VA_2weeks: Visual acuity (measured in ETDRS letters) of the patient 2 weeks after surgery.
• VA_3months: Visual acuity (measured in ETDRS letters) of the patient 3 months after surgery.
• VA_6months: Visual acuity (measured in ETDRS letters) of the patient 6 months after surgery.
• VA_12months: Visual acuity (measured in ETDRS letters) of the patient 12 months after surgery.

References

Godbout, M., et al. "Predicting Visual Improvement after Macular Hole Surgery: a Cautionary Tale on Deep Learning with Very Limited Data." arXiv preprint arXiv:2109.09463 (2021).