Development of pectoralis muscle segmentation method from chest CT image using 2.5-dimensional multichannel image dataset

Congress:

ECR 2020

Poster Number:

C-02697

Type:

Scientific Exhibit

Keywords:

Artificial Intelligence and Machine Learning, Artificial Intelligence, Lung, Musculoskeletal soft tissue, CAD, CT, Computer Applications-Detection, diagnosis, Segmentation, Chronic obstructive airways disease, Retrospective, Not applicable, Performed at one institution

Authors:

Z. Yang, H. Lee, T. Choi, J. Jung, M. Ryu, H. Yong; Seoul/KR

DOI:

10.26044/ecr2020/C-02697

DOI-Link:

https://dx.doi.org/10.26044/ecr2020/C-02697

Fig. 1: The overall process of the proposed method.

Fig. 2: (a) graphical display for the sagittal image extraction process. (b) an example...

Fig. 6: example images of proposed segmentation method.

Table 1: The correlation result between COPD diagnostic biomarkers with measured...

Purpose

Chronic obstructive pulmonary disease (COPD) has been described as a systemic disease[1–6], which is known to be highly related to various adverse outcomes. The range of outcome is not only parenchymal function related disease, but also respiratory muscle-related diseases such as air trapping, dyspnea, reducing oxygen utilization, skeletal muscle dysfunction, and sarcopenia. Pectoralis major muscle is one of the major respiratory muscle, covering most of the chest part and connecting the shoulder and abdomen. Many studies showed that the density and the volume of pectoralis...

Methods and materials

The segmentation method was developed based on deep-learning techniques with combining muscle area detection model and segmentation model, for solving the weight-imbalance problem in training dataset, and for getting better process efficiency of the deep learning model. For training and validation of the model, 200 patient CT studies were obtained from Korea University Guro Hospital. Among them, 90 patients were COPD patients, while rest 110 patients were normal subjects from the diagnostic cohort. All CT examinations were performed using a diagnostic routine with normal dose...

Results

In pectoralis range detection, the training accuracy and the loss value of the validation set were 0.9954 and 0.0725, respectively. And in segmentation, the value were 0.9836 and 0.1725, respectively. Visual assessment results were scored on 9.3 from pectoralis muscle range detection, and 8.7 on overall masks segmentation result. The correlation with pectoralis muscle volume was shown relatively high on VC, FEV1, DLCO, and TLC, which is shown in table 1.

Conclusion

Deep learning-based pectoralis muscle segmentation method showed appropriate segmentation result, which can be used for measuring the pectoralis muscle volume, mass, density for further research.

Personal information and conflict of interest

Z. Yang; Seoul/KR - nothing to disclose H. Lee; Seoul/KR - nothing to disclose J. Jung; Seoul/KR - nothing to disclose M. Ryu; Seoul/KR - nothing to disclose T. Choi; Seoul/KR - nothing to disclose H. Yong; Seoul/KR - nothing to disclose

References

[1] Wouters EFM, Creutzberg EC, Schols AMWJ. Systemic effects in COPD. Chest, vol. 121, American College of Chest Physicians; 2002, p. 127S-130S. https://doi.org/10.1378/chest.121.5_suppl.127S. [2] Andreassen H, Vestbo J. Chronic obstructive pulmonary disease as a systemic disease: An epidemiological perspective. Eur Respir Journal, Suppl 2003;22. https://doi.org/10.1183/09031936.03.00000203. [3] Celli BR, Cote CG, Marin JM, Casanova C, Montes De Oca M, Mendez RA, et al. The Body-Mass Index, Airflow Obstruction, Dyspnea, and Exercise Capacity Index in Chronic Obstructive Pulmonary Disease. N Engl J Med 2004;350:1005–12. https://doi.org/10.1056/NEJMoa021322. [4] Schols...