|
--- |
|
license: cc-by-nc-4.0 |
|
task_categories: |
|
- feature-extraction |
|
tags: |
|
- medical |
|
size_categories: |
|
- 1M<n<10M |
|
--- |
|
 |
|
|
|
<div align="center"> |
|
<h1>Surgical Youtube Dataset</h1> |
|
|
|
Welcome to the **Surgical YouTube Dataset!** This curated collection of surgical video frames is a key contribution of our research, playing a role in |
|
enhancing the performance of our foundational model, **SurgeNetXL**. |
|
|
|
The dataset is derived from **680 hours** of surgical video footage obtained from YouTube, processed using the code |
|
from [Schmidgall et al. (2024)](https://github.com/SamuelSchmidgall/GSViT) and sampled at **1 frame per second (fps)**. Through manual annotation, |
|
we ensured the removal of all non-minimally invasive procedures and out-of-body frames, resulting in a dataset comprising **2,074,234 frames** across **23 distinct surgical procedures**. |
|
|
|
The **Surgical YouTube Dataset** is publicly available and serves as a key component of our **SurgeNetXL** model, |
|
which also integrates several other open-source datasets. These additional datasets are detailed in the table below. |
|
|
|
Models trained on this dataset, including **SurgeNetXL**, can be found on [github](https://github.com/TimJaspers0801/SurgeNet). |
|
|
|
|
|
| Procedure-specific subset | Dataset | Procedure | #videos | #frames | Public | |
|
|---------------------------|----------------------------------------------------------------|-----------|---------|-----------|--------| |
|
| **SurgeNetCholec** | Cholec80 ([Twinnanda et al., 2017b](https://arxiv.org/abs/1602.03012)) | Laparoscopic Cholecystectomy | 76 | 179,164 | Yes | |
|
| | HeiChole ([Maier-Hein et al., 2021](https://www.synapse.org/Synapse:syn25101790/wiki/608802)) | Laparoscopic Cholecystectomy | 30 | 53,427 | Yes | |
|
| | hSDB-Chole ([Yoon et al., 2021](https://arxiv.org/abs/2110.12555)) | Laparoscopic Cholecystectomy | 24 | 18,064 | Yes | |
|
| **SurgeNetRAMIE** | RAMIE-UMCU | RA Esophagectomy | 28 | 377,287 | No | |
|
| **SurgeNetRARP** | ESAD [Bawa et al., 2021](https://arxiv.org/abs/2006.07164) | RA Esophagectomy | 28 | 47,282 | Yes | |
|
| | PSI-AVA [Valderrama et al., 2022](https://arxiv.org/abs/2212.04582) | RA Prostatectomy | 8 | 73,618 | Yes | |
|
| | RARP-AvL | RA Prostatectomy | 8 | 261,516 | No | |
|
| **Others** | DSAD ([Carstens et al., 2023](https://www.nature.com/articles/s41597-022-01719-2)) | RA Rectal Resection/Extirpation | 32 | 14,623 | Yes | |
|
| | GLENDA ([Leibetseder et al., 2020](https://link.springer.com/chapter/10.1007/978-3-030-37734-2_36)) | Gynecologic Laparoscopy | 400 | 25,682 | Yes | |
|
| | LapGyn4 ([Leibetseder et al., 2018](https://dl.acm.org/doi/10.1145/3204949.3208127)) | Gynecologic Laparoscopy | 500 | 59,616 | Yes | |
|
| | MultiBypass140 ([Lavanchy et al., 2024](https://github.com/CAMMA-public/MultiBypass140)) | Laparoscopic Gastric Bypass Surgery | 140 | 749,419 | Yes | |
|
| | hSDB-Gastric ([Yoon et al., 2021](https://arxiv.org/abs/2110.12555)) | RA Gastrectomy | 24 | 35,576 | Yes | |
|
| | SurgToolLoc2022 ([Zia et al., 2023](https://arxiv.org/abs/2305.07152)) | 11 different RA porcine procedures | N/A | 741,516 | Yes | |
|
| | YouTube [ours](https://huggingface.co/datasets/TimJaspersTue/SurgeNetYoutube) | 23 identified procedures | 3,253 | 2,074,234 | Yes | |
|
|
|
| SurgeNetXL variations | Dataset | Procedure | #videos | #frames | Public | |
|
|-----------------------|------------------------------------------------------------|---------------------------------------------------------|---------|---------|--------| |
|
| **SurgeNetSmall** | 10% of the above (excluding YouTube) | All of the above (excluding YouTube) | \>1345 | 263,679 | Partly | |
|
| **SurgeNetPublic** | All public datasets (excluding YouTube & private datasets) | All of the above (excluding YouTube & RA Esophagectomy) | \>1238 | 1,997,987 | Yes | |
|
| **SurgeNet** | All of the above (excluding YouTube) | All of the above (excluding YouTube) | \>1345 | 2,636,790 | Partly | |
|
| **SurgeNetXL** | All of the above | All of the above | \>4598 | 4,711,024 | Partly | |
|
|
|
|
|
<h1>Publication</h1> |
|
|
|
<h3>Scaling up self-supervised learning for improved surgical foundation models</h3> |
|
|
|
[Tim J.M. Jaspers](https://timjaspers0801.github.io/)<sup>1* :email:</sup>, [Ronald L.P.D. de Jong](https://scholar.google.com/citations?user=We226GgAAAAJ&hl=en)<sup>2*</sup>, |
|
[Yiping Li](https://research.tue.nl/nl/persons/yiping-li/publications/)<sup>2</sup>, [Carolus H.J. Kusters](https://chjkusters.github.io/)<sup>1</sup>, Franciscus H.A. Bakker<sup>5</sup>, |
|
Romy C. van Jaarsveld<sup>3</sup>, Gino M. Kuipers<sup>3</sup>, Richard<sup>3</sup>, Jelle P. Ruurda<sup>3</sup>, |
|
Willem M. Brinkman<sup>4</sup>, Josien P.W. Pluim<sup>2</sup>, |
|
Peter H.N. de With<sup>1</sup>, Marcel Breeuwer<sup>2</sup>, [Yasmina Al Khalil](https://scholar.google.com/citations?user=m6co7N0AAAAJ&hl=en)<sup>2</sup>, [Fons van der Sommen](https://scholar.google.com/citations?user=qFiLkCAAAAAJ&hl=en)<sup>1</sup> |
|
|
|
<sup>1</sup> Department of Electrical Engineering, Video Coding & Architectures, Eindhoven University of Technology \ |
|
<sup>2</sup> Department of Biomedical Engineering, Medical Image Analysis, Eindhoven University of Technology, Eindhoven, The Netherlands \ |
|
<sup>3</sup> Department of Surgery, University Medical Center Utrecht, Utrecht, The Netherlands \ |
|
<sup>4</sup> Department of Oncological Urology, University Medical Center Utrecht, Utrecht, The Netherlands \ |
|
<sup>5</sup> Department of Surgery, Catharina Hospital, Eindhoven, The Netherlands |
|
|
|
<sup>*</sup> Both authors attributed equally \ |
|
(<sup>:email:</sup>) corresponding author |
|
|
|
*arxiv* <br /> ([Article](https://arxiv.org/abs/2501.09436)) |
|
|
|
|
|
<h1>Abstract</h1> |
|
Foundation models have revolutionized computer vision by achieving state-of-the-art |
|
performance across diverse tasks through large-scale pretraining on extensive datasets. |
|
However, their application in surgical computer vision has been limited. This study |
|
addresses this gap by introducing SurgeNetXL, a novel surgical foundation model |
|
that sets a new benchmark in surgical computer vision. Trained on the largest reported surgical dataset to date, |
|
comprising over 4.7 million video frames, SurgeNetXL |
|
achieves consistent top-tier performance across six datasets spanning four surgical |
|
procedures and three tasks, including semantic segmentation, phase recognition, and |
|
critical view of safety (CVS) classification. Compared to the best-performing surgical |
|
foundation models, SurgeNetXL shows mean improvements of 0.26%, 8.95%, |
|
and 12.6% for semantic segmentation, phase recognition, and CVS classification, respectively. |
|
Additionally, SurgeNetXL outperforms the best-performing ImageNet-based variants |
|
by 10.3%, 4.0%, and 1.6% in the respective tasks. In addition to |
|
advancing model performance, this work provides key insights into scaling pretraining datasets, |
|
extending training durations, and optimizing model architectures specifically for surgical computer vision. |
|
These findings pave the way for improved generalizability and robustness in data-scarce scenarios, |
|
offering a comprehensive framework for future research in this domain. |
|
|
|
|
|
<h1>Results</h1> |
|
The following figures are from our publication, showcasing the performance of our introduced foundation model |
|
across diverse surgical tasks and procedures. These results demonstrate the model’s state-of-the-art |
|
performance on a variety of downstream tasks, reflecting its versatility and robustness in handling |
|
datasets from multiple surgical procedures. |
|
|
|
Figure 1 and Figure 2 illustrate comparative rankings of our model against existing benchmarks, |
|
highlighting its superior generalization capabilities across datasets. Figure 3 provides a t-SNE visualization, |
|
showcasing the clear cluster separation per specific dataset achieved by the model’s feature embeddings, |
|
further emphasizing its effectiveness in capturing meaningful representations. |
|
|
|
<div style="display: flex; justify-content: space-around; align-items: center; gap: 20px;"> |
|
|
|
<div style="text-align: center;"> |
|
<img src="figures/radar_ranks.png" alt="Fig 2" width="400" height="300"> |
|
<p><strong>Fig 1:</strong> Radar chart showing model ranks across datasets.</p> |
|
</div> |
|
|
|
<div style="text-align: center;"> |
|
<img src="figures/ranking_blob_all.png" alt="Fig 3" width="400" height="300"> |
|
<p><strong>Fig 2:</strong> Blob chart representing ranking metrics for models.</p> |
|
</div> |
|
|
|
</div> |
|
|
|
<div style="text-align: center"> |
|
<img src="figures/TSNE.png" alt="Fig 3" width="800"> |
|
<p><strong>Fig 3:</strong> t-SNE visualization of feature embeddings showing cluster separation across datasets.</p> |
|
</div> |
|
|
|
|
|
<h1>Citation</h1> |
|
If you use this Surgical Youtube dataset or one of our models please cite our paper: |
|
</div> |
|
|
|
```bibtex |
|
@misc{jaspers2025scalingselfsupervisedlearningimproved, |
|
title={Scaling up self-supervised learning for improved surgical foundation models}, |
|
author={Tim J. M. Jaspers and Ronald L. P. D. de Jong and Yiping Li and Carolus H. J. Kusters and Franciscus H. A. Bakker and Romy C. van Jaarsveld and Gino M. Kuiper and Richard van Hillegersberg and Jelle P. Ruurda and Willem M. Brinkman and Josien P. W. Pluim and Peter H. N. de With and Marcel Breeuwer and Yasmina Al Khalil and Fons van der Sommen}, |
|
year={2025}, |
|
eprint={2501.09436}, |
|
archivePrefix={arXiv}, |
|
primaryClass={cs.CV}, |
|
url={https://arxiv.org/abs/2501.09436}, |
|
} |
|
``` |
