A new 3-dimesional device positioning method using C-arm X-ray fluoroscopy for transbronchial biopsy- A study of error estimation using machine learning -

Congress:

ECR 2020

Poster Number:

C-05266

Type:

Scientific Exhibit

Keywords:

Performed at one institution, Not applicable, Retrospective, Cancer, Computer Applications-General, Biopsy, Fluoroscopy, Lung, Projection Radiography

Authors:

C. Nagashima, M. kitagawa, N. Ikeno, Y. Matsumoto, M. Tanaka, T. IIMURA, K. Matsuzaki, K. Ihara, T. Aso; Tokyo/JP

DOI:

10.26044/ecr2020/C-05266

DOI-Link:

https://dx.doi.org/10.26044/ecr2020/C-05266

Fig. 1: 3D device positioning method.

Fig. 2: An example image of point error by 3D device positioning method.

Fig. 3: Map of error between the point indicated by the 3D device positioning method...

Fig. 4: Results of predicted value of error using machine learning: Multiple liner...

Purpose

Usually, the position of the biopsy device in bronchoscopy for peripheral pulmonary lesions（PPLs） is confirmed by X-ray fluoroscopy. However, X-ray image has been not easy to confirm the position in the anterior-posterior (AP) direction, and the location of the device might not be recognized well in some cases. Therefore, we developed a 3-dimensional (3D) device positioning method that recognizes the position of devices in transbronchial biopsy combining preprocedural computed tomography (CT) and periprocedural X-ray fluoroscopy images (Figure 1). Before, we reported well accuracy of the...

Methods and materials

Patient Population We analyzed 138 clinical cases from transbronchial biopsy for PPLs between November 2016 and September 2017. All cases of transbronchial biopsy were performed using radial endobronchial ultrasound (R-EBUS) and C-arm X-ray fluoroscopy (VersiFlex VISTA, Hitachi Ltd.). We selected cases acquired three directions X-ray images (RAO 45 °, 0 °, LAO 45 °) for confirm device location. Lesions with a long diameter is 3 cm or longer were excluded to difficult verifying device location. In addition, usually doctors observe lesion using R-EBUS and determine...

Results

Verification of error by 3D device positioning method The median error was 18.0 mm (range: 2.7-72.2 mm). Figure 3 shows error between the point indicated by the 3D device positioning method and the pointed out by the bronchoscopist. A certain tendency was observed in size and direction of the error in each lobe region. The error direction was observed mainly from superior to inferior and from posterior to anterior. The error size was larger lower lobe than upper and middle lobe (Table 1). We estimated...

Conclusion

We evaluated the analysis error of the 3D device positioning method and performed the error estimation by machine learning. We estimated that the error of our 3D device positioning method was mainly due to respiratory movements. In order to correct the error due to the respiratory movement, we tried to estimate the error by machine learning. The machine learning using random forest regression showed the best performance in this study. We plan to conduct the study to compensate the error applying this technique.

Personal information and conflict of interest

C. Nagashima; Tokyo/JP - Research/Grant Support at Hitachi Ltd. T. Aso; Tokyo/JP - Research/Grant Support at Hitachi Ltd. K. Ihara; Tokyo/JP - Research/Grant Support at Hitachi Ltd. M. Kitagawa; Tokyo/JP - Research/Grant Support at Hitachi Ltd. Y. Matsumoto; Tokyo/JP - Research/Grant Support at Hitachi Ltd. M. Tanaka; Tokyo/JP - Research/Grant Support at Hitachi Ltd. N. Ikeno; Tokyo/JP - Research/Grant Support at Hitachi Ltd. T. Iimura; Tokyo/JP - Employee at Hitachi Ltd. K. Matsuzaki; Tokyo/JP - Employee at Hitachi Ltd.

References

[1] M. Kitagawa et al. ECR2018 C-2401 [2] Y. Tong et al. (2017) Chest Fat Quantification via CT Based on Standardized Anatomy Space in Adult Lung Transplant Candidates. PLOS ONE. 12(1): e0168932