About

Frozen three feet is not a day's cold
Making mountains from mounds is not done in a short time

Hi, I am Shuai Lei，Both my master's and master's degrees are in Hohai University. At present, I am a research junior student in the School of Information Engineering, Hohai University.

I primarily worked under the guidance of Prof. JunFeng Jiang on medical image processing. I participated in several projects and tasks at the ChangZhou Key Laboratory of Digital Technology on Graphics and Orthopaedic implants. My work involved object detection、image segmentation、keypoints recognition and 3D/2D registration.

My github home page👉GitHub·ThreeStones and CSDN home page👉SL1029。

News

• 2024.10.14 The paper《ABLSpineLevelCheck: Localization of Vertebral Levels on Fluoroscopy via Semi-supervised Abductive Learning》was accepted by BIBM2024 as a short paper(Acceptance rate:21%).

Main Projects

• 2024.04-now Contrastive Learning for Fracture recognition in Intraoperative Radiographs
• 2023.08-2024.04 Vertebrae Recognition on Intraoperative Radiographs Using Abductive Learning(BIBM2024)


Deep learning has demonstrated promising efficacy in the localization of vertebrae within X-ray imagery, although it is recognized for its deficiencies in compositional generalization, data efficiency, and interpretability. To address this issue, we introduce an abductive learning mechanism, situated within the neuro-symbolic paradigm, tailored for semi-supervised vertebral localization. Initially, unannotated spinal fluoroscopic images are processed by the networks to infer pseudo-labels for vertebra localization. Subsequently, these pseudo-labels undergo abductive reasoning via a knowledge base comprised of first-order logical clauses. The networks are then retrained utilizing the abducted outcomes. Additionally, we propose an ensemble technique that amalgamates semantic detection of vertebral levels with instance detection. To further augment performance, we have synthesized a dataset and annotated the BUU dataset for network pretraining. Ablation studies validate the efficacy of the proposed components in our methodology. Furthermore, comparative analyses reveal that our approach significantly surpasses leading object detection algorithms, exhibiting superior performance with minimal annotations.

• 2023.02-2023.07 Vertebral segmentation and localization in spine CT

Other Projects

• 2024.01 Spine DRR generation and annotation
python code 和 C++ code

The ITK projection is used to obtain the horizontal rectangular box and the rotating rectangular box according to the boundary of the mask. At the same time, the projected mask is obtained as the segmentation label.
• 2023.11-2023.12 Vertebra segmentation based on nnunet

Firstly, the semantic recognition of the vertebrae is performed, and each vertebra is cut out according to the detection box. Then nnunet is used to perform 0,1 segmentation to obtain the segmentation results of each segment.
• 2023.07-2023.08 Keypoint detection of vertebrae based on HRNet
Data annotation

Data annotation was performed by open3d to project four points on the vertebral body, including the center of the upper plane of the vertebral body, the center of the lower plane of the vertebral body, and the left and right pedicles, onto the DRR. Results