Histological Dataset for Microvascular Segmentation of Tissue-Engineered Vascular Grafts

Objectives: The pursuit of understanding vascular tissue regeneration within tissue-engineered vascular grafts (TEVGs) is of paramount importance due to the critical role these grafts play in replacing damaged or diseased blood vessels. TEVGs offer a promising alternative to traditional grafts, with the potential to integrate into the host's tissue and support the natural regenerative processes. However, challenges such as thrombosis, inflammation, and the need for grafts that can adapt to the dynamic biological environment remain. By studying the regenerative processes in TEVGs, researchers can gain insights into the mechanisms that underpin successful graft integration and function, which is essential for improving patient outcomes in vascular surgeries. This dataset, with its detailed annotations of histological features, provides a valuable resource for developing and refining machine-learning models that can analyze and predict patterns of vascular tissue regeneration. The ability to accurately segment and quantify microvessels and immune cells in regenerated arteries is a significant step forward in distinguishing between physiological and pathological regeneration, ultimately contributing to the design of more effective and reliable TEVGs for clinical use.

Ethical Approval: Experimental strategy of the study is described in detail in [1] and [2]. The study was conducted according to the guidelines of the Declaration of Helsinki, and was approved by the Local Ethical Committee of the Research Institute for Complex Issues of Cardiovascular Diseases (Kemerovo, Russia, protocol code 2020/06, date of approval: 19 February 2020). Animal experiments were performed in accordance with the European Convention for the Protection of Vertebrate Animals (Strasbourg, 1986) and Directive 2010/63/EU of the European Parliament on the protection of animals used for scientific purposes. For the implantation, we used female Edilbay sheep of 42–45 kg body weight which were received from the Animal Core Facility of the Research Institute for Complex Issues of Cardiovascular Diseases (Kemerovo, Russia) and selected for the surgery by Doppler ultrasonography to identify those having carotid artery diameter of 4.0 ± 0.2 mm.

Description: The dataset comprises a collection of Whole Slide Images (WSIs) obtained from biodegradable TEVGs implanted into the carotid arteries of 20 sheep. A total of 104 WSIs were acquired, each measuring an average size of 135,000 x 123,000 pixels. These WSIs were stained using Hematoxylin and Eosin (H&E), a common practice for highlighting the structure of tissue sections, which facilitates the detailed examination of histological features. These WSIs were automatically sliced into 99,831 patches of 3,000 x 3,000 pixels and subsequently filtered, resulting in 1,401 selected patches for manual annotation.

Annotation Method: Two pathologists independently selected and meticulously annotated the 1401 patches, identifying nine distinct histological features associated with vascular tissue regeneration. These features include arteriole lumen (AL), arteriole media (AM), arteriole adventitia (AA), venule lumen (VL), venule wall (VW), capillary lumen (CL), capillary wall (CW), immune cells (IC), and nerve trunks (NT). The annotations were performed using binary masks, delineating each feature within the patches. Subsequently, a senior pathologist conducted a triple verification process, reviewing and refining the annotations to ensure accuracy and consistency. The annotations are provided in the form of binary masks, meticulously defined for each feature within the patches.

Dataset Split: Given the limited number of subjects studied, comprising 20 sheep, we employed a 5-fold cross-validation technique to split our dataset. This method was chosen because it allows for the efficient use of limited data, ensuring that each observation has the opportunity to be used in both the training and testing sets, thus reducing bias and providing a more accurate estimate of the model's performance. In this approach, each fold involved 16 sheep for training and the remaining 4 for testing (see Table 1 and Figure 3). This partitioning scheme was consistently applied to maintain the integrity of subject groups within each subset and to prevent data leakage. The 5-fold cross-validation is particularly beneficial for our study's objectives as it maximizes the training data available for developing robust machine learning models while also ensuring that the models are tested on unseen data, thereby enhancing the generalizability of our findings.

Access to the Study: Further information about this study, including curated source code, dataset details, and trained models, can be accessed through the following repositories:

Source code: https://github.com/ViacheslavDanilov/histology_segmentation

Dataset: https://doi.org/10.5281/zenodo.10838384

Models: https://doi.org/10.5281/zenodo.10838431

Table 1. Patch and feature distributions across folds and subsets

Fold

Subset

Patches

AL

AM

AA

VL

VW

CL

CW

IC

NT

Total

1

Train

1168

510

512

220

675

648

770

765

409

448

4957

1

Test

233

81

84

36

186

169

178

182

91

25

1032

2

Train

1053

406

411

179

678

638

743

746

423

315

4539

2

Test

348

185

185

77

183

179

205

201

77

158

1450

3

Train

1127

507

511

222

743

702

759

760

299

423

4926

3

Test

274

84

85

34

118

115

189

187

201

50

1063

4

Train

1064

466

472

199

611

566

759

758

423

291

4545

4

Test

337

125

124

57

250

251

189

189

77

182

1444

5

Train

1192

475

478

204

737

714

761

759

446

415

4989

5

Test

209

116

118

52

124

103

187

188

54

58

1000