Advancing Vascular Health in Athletes: Application of Deep Fully Convolutional Neural Networks for Enhanced Imaging and Diagnosis

Authors

  • Xinrui Xu Shanxi Key Laboratory of Birth Defect and Cell Regeneration, Department of Biochemistry and Molecular Biology, School of Basic Medical Science, Shanxi Medical University, Taiyuan 030001, China
  • Jun Xie Shanxi Key Laboratory of Birth Defect and Cell Regeneration, Department of Biochemistry and Molecular Biology, School of Basic Medical Science, Shanxi Medical University, Taiyuan 030001, China.
  • Yuchen Wang School of Computer Science and Engineering, University of Electronic Science and Technology of China, Chengdu 611731, China
  • Zixuan Hu Department of Physiology and Pathophysiology, School of Basic Medical Science, Shanxi Medical University, Taiyuan 030001, China
  • Zhiwei Peng Shanxi Key Laboratory of Birth Defect and Cell Regeneration, Department of Biochemistry and Molecular Biology, School of Basic Medical Science, Shanxi Medical University, Taiyuan 030001, China
  • Yuxiang Liang Shanxi Key Laboratory of Birth Defect and Cell Regeneration, Department of Biochemistry and Molecular Biology, School of Basic Medical Science, Shanxi Medical University, Taiyuan 030001, China
  • Zhizhen Liu Shanxi Key Laboratory of Birth Defect and Cell Regeneration, Department of Biochemistry and Molecular Biology, School of Basic Medical Science, Shanxi Medical University, Taiyuan 030001, China
  • Hong Zhao Shanxi Key Laboratory of Birth Defect and Cell Regeneration, Department of Biochemistry and Molecular Biology, School of Basic Medical Science, Shanxi Medical University, Taiyuan 030001, China
  • Qizhi Shuai Shanxi Key Laboratory of Birth Defect and Cell Regeneration, Department of Biochemistry and Molecular Biology, School of Basic Medical Science, Shanxi Medical University, Taiyuan 030001, China

Keywords:

Tumor vascular Mimicry; Deep Fully Convolutional Neural Network; Image Segmentation

Abstract

Background: In the realm of sports medicine, the early detection and management of vascular conditions are critical, particularly in sports where high physical demand can lead to vascular stress and anomalies that may exhibit patterns similar to tumor vascular mimicry (VM). VM plays a crucial role in tumor growth, metastasis, and resistance, posing significant challenges in imaging due to its scarcity and complex contours in pathological tissues. Method: We adapted a fully convolutional neural network model, originally developed for VM segmentation in cancer biopsies, to potentially identify and analyze similar vascular patterns in athletes. This model utilizes a two-stage process: down sampling and up sampling. In the down sampling stage, convolutional layers from ResNet-34 pretrained on ImageNet-1k are employed to create high-dimensional feature maps of the target images. In the up sampling stage, transposed convolution is used to revert the feature map back to the size of the original image. Further refinement through transposed convolution mid-network optimizes the output. Results: Applied to a dataset of 212 biopsy images potentially containing VM, the model demonstrated a pixel accuracy of 98.38% on the test set. The high accuracy underscores the model’s capability in identifying and quantifying complex vascular patterns, suggesting its applicability in sports medicine for evaluating vascular health in athletes. Conclusion: The adapted deep fully convolutional network model offers a promising tool for advanced imaging and analysis of vascular conditions in athletes. By leveraging this model, sports medicine professionals can achieve greater precision in diagnosing and monitoring vascular health, potentially improving outcomes for athletes facing high-risk vascular conditions. Further research is needed to validate and refine this application, with a focus on tailoring the approach to the specific vascular challenges faced in competitive sports.

Downloads

Published

2024-07-01

Issue

Vol. 23 No. 89.1 (2023): Vol. 23, No. 89.1, abril/April 2023

Section

Articles