High-Quality Lung 3D Reconstruction Dataset and Model Sharing

[lirongyaoper (Rongyao Li)]

Here is the professional translation of your text into English:

---

High-Quality Lung 3D Reconstruction Dataset and Model Sharing

1. Dataset Information
I possess high-quality annotated lung 3D reconstruction datasets for model training. Both raw data and labels are formatted according to nnUNetV2 standards and are ready for immediate training without additional preprocessing.

Special Note:
If you also have high-quality lung annotation data, we are open to 1:1 data exchange.
Exchange Principles:

1. Exchanged data must meet high-quality standards.
2. No duplication between exchanged datasets.
3. Duplicate data (identical content under different filenames) is strictly prohibited.
   For exchange details, please contact me privately.

---

2. Pretrained Model Sharing
A lung vasculature and airway segmentation model with Dice score 0.88 is now publicly available. Researchers and clinicians may use this model for preoperative reconstruction.

Important Notes:

• Due to limited training data, minor inaccuracies may exist. Users should carefully validate critical anatomical details.
• Feedback and suggestions for improvement are strongly encouraged to help develop an open-source model comparable to commercial solutions.
• Technical support is available via email: lryhdf@163.com

链接: https://pan.baidu.com/s/15qQYK-LsSZAiwL69IzdUdg 提取码: y728
–来自百度网盘超级会员v9的分享

3. Recommended Hardware Specifications
For optimal local 3D lung reconstruction performance, we recommend the following configuration:

Component	Specification
CPU	Intel i7-14700K or equivalent
RAM	≥48GB
GPU	NVIDIA RTX 4070 Ti or higher
VRAM	≥16GB (24GB+ preferred)
Storage	NVMe SSD with adequate capacity

Other components should be configured to match this performance tier.

---

This translation maintains technical precision while ensuring clarity for international collaborators. Let me know if further refinements are needed!

[pieper (Steve Pieper (Isomics, Inc.))]

Thanks for posting about this, your segmentations look very nice.

You mentioned that the segmentation model is publicly available, can you provide details? I don’t know if it was meant to be at the baidu link you shared, but unfortunately that link did not work for me.

It would be great if we could try your model on some of our datasets and possibly integrate it in Slicer for anyone to try.

[mau_igna_06 (Mauro I. Dominguez)]

Very nice!

@pieper, you need to enter the download code: y728

And because the file is large, to download it you need an account on that website

[lirongyaoper (Rongyao Li)]

Dear Professors,

The pulmonary vascular and airway segmentation model I’m sharing was trained on my local computer using the nnUnetV2 framework with 1,522 standardized clinical cases. Due to limited local computing resources, I have currently only completed 5-fold training in 3D full-resolution configuration, achieving a final Dice score of 0.88. There remains a gap compared to commercial models. In light of this, I would like to clarify:

1. The segmentation model I share is intended for clinical reference only and must not be used for commercial purposes.

2. I would appreciate the opportunity to obtain additional standardized datasets to further improve the model through continued training, and would be grateful if you could consider sharing relevant data with me.

3. If any professors possess advanced GPU server resources and more sophisticated training frameworks, and would be willing to share the resulting trained models with me, I am prepared to provide my original standardized dataset to support further collaborative training efforts.