A preliminary examination of the diagnostic value of deep learning in stage of thymoma

Congress:

ECR 2019

Poster Number:

C-1202

Type:

Scientific Exhibit

Keywords:

Artificial Intelligence, Computer applications, Breast, CAD, CT, Neural networks, Computer Applications-Detection, diagnosis, Screening

Authors:

X. Tian¹, R. Zhang¹, C. Xia¹, S. Liang¹, L. Yang², Y. Zhu²; ¹Beijing/CN, ²Guangzhou/CN

DOI:

10.26044/ecr2019/C-1202

DOI-Link:

https://dx.doi.org/10.26044/ecr2019/C-1202

Fig. 2: Dense Block by G. Huang et al. [3]

Fig. 1: Prediction procedure of thymoma stages, derived from G. Huang et al. [3]

Fig. 3: Experiment results

Aims and objectives

Thymoma is a type of common anterior mediastinal tumor. Evaluation of progress of thymoma is widely based on Masaoka-Koga staging system [1, 2]. Masaoka-Koga based classification for stage I from other stages accords to whether the thymoma breaks the tumor capsule or not. The invasion of tumor capsule is an important factor related to prognosis. Thus, classification of stage I from stage II to IV is of great significance for patients' survival. However, it is not easy for radiologists to directly stage thymoma from images,...

Methods and materials

Our model predicts the stages based on the ROI (region of interests) from CT images, and a modified DenseNet [3] is designed to extract the 3D features of ROI for classification. Data Description: The thymoma dataset has total number of 184. The training set consists of 147 chest CT sequences (80%), and the test dataset has 37 chest CT sequences (20%). The gold standard was acquired from institutional database for medical records which identified patients with histologically confirmed thymic epithelial tumor who underwent surgical resection,...

Results

Weused five-fold stratified cross validation in this experiment. The average accuracy of our model achieved 74.8% in training set, and 68.7% in test set. By contrast, the average accuracy by 3 chest-radiologists was 60.7% (Fig.3).

Conclusion

The results showed that our deep learning model obtained high accuracy in pre-operative staging for thymoma, which can reach the similar radiologist performance on thymoma pre-operative staging.

Personal information

X. Tian Beijing InferVision Technology Co., Ltd. Block A, Yuanyang International Center, East Fourth Ring Road, Chaoyang District, Beijing, China 100025 Phone: +86 18600863828 E-mail: [email protected]

References

[1]Masaoka A, Monden Y, Nakahara K, et al. Follow-up study of thymomas with special reference to their clinical stages. Cancer 1981;48:2485-92. [2] Koga K, Matsuno Y, Noguchi M, et al. A review of 79 thymomas: modification of staging system and reappraisal of conventional division into invasive and non-invasive thymoma. Pathol Int 1994;44:359-67. [3] G. Huang, Z. Liu, L. Van der Maaten, and K. Q. Weinberger. Densely connected convolutional networks. In CVPR, 2017. 2.2.1, 2.11 [4] L. Fei-Fei, ImageNet: crowdsourcing, benchmarking & other cool things, CMU...