Labeling Human Embryo Time-Lapse Videos: A Dataset Refinement Approach

Introduction

This dataset contains microscopic images of human embryos captured through time-lapse videos. It is an evolution of the dataset originally created from the study published by Gomez Tristan, et al. However, unlike the original study, this dataset focuses exclusively on zygotes—the initial stage of cellular maturation, between fertilization and the first cleavage. The original dataset consists of seven layers of images, each representing a microscopic focal plane of the same embryo. This multidimensional approach allows for a more precise visualization of cellular details, making it highly relevant for dataset creation. However, working with multiple layers is not the most accessible format for most computer vision technologies. Additionally, although the original dataset includes temporal markings for each stage, it does not provide object localization within the images. Therefore, this dataset enhances the previous version by providing annotations for the position of key morphological features of the zygote stage: pronuclei, polar bodies, vacuoles, and cytoplasmic granules. Furthermore, it reduces the structural complexity of the original dataset, adapting the images into a more accessible format compatible with approaches using up to three channels (RGB).

Class:

To zygote_non_zygote_classification:

Class 0: zygote

Class 1: non zygote

To zygote_artefacts_detection:

Class 0: pronucleus

Class 1: cytoplasmic granules

Class 2: polar body

Class 3: vacuole

Preprocessing:

Center Layer: Only the F0 layer.

Three Layers: An RGB imagen using F+15, F0 and F-15.

Full Flattened: A flattened image created using Focus Stacking with all layers.

Division

dataset/│├─ images/│  ├─ zygote_artefacts_detection/│  │  ├─ center_layer/ [test, train, val]│  │  ├─ full_flatted/ [test, train, val]│  │  └─ three_layers/ [test, train, val]│  ││  └─ zygote_non_zygote_classification/│     ├─ center_layer/ [test, train, val]│     ├─ full_flatted/ [test, train, val]│     └─ three_layers/ [test, train, val]│└─ labels/   ├─ zygote_artefacts_detection/ [test, train, val]   └─ zygote_non_zygote_classification/ [test, train, val]

Counting and dividing data

Images

Labels

Cellular Structures

Pronucleus

granules

polar body

vacuole

Train

6501

75%

12063

6180

13819

2523

110

Validation

1657

19%

3273

1227

2718

1032

30

Test

510

6%

579

432

947

99

34

Total

8668

100%

15915

7839

17484

3654

174

Images

Zygote labels

Non-zygote labels

Cellular Structures

Train

8562

71%

3516

5046

489

Validation

1809

15%

787

1022

100

Test

1628

14%

692

936

100

Total

11999

100%

4.995

7.004

689