Two-stage classification of tuberculosis culture diagnosis using convolutional neural network with transfer learning
- PDF / 1,863,621 Bytes
- 16 Pages / 439.37 x 666.142 pts Page_size
- 93 Downloads / 207 Views
Two‑stage classification of tuberculosis culture diagnosis using convolutional neural network with transfer learning Ray‑I Chang1 · Yu‑Hsuan Chiu1 · Jeng‑Wei Lin2
© Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract Tuberculosis (TB) has been one of top 10 leading causes of death. A computer-aided diagnosis system to accelerate TB diagnosis is crucial. In this paper, we apply convolutional neural network and deep learning to classify the images of TB culture test—the gold standard of TB diagnostic test. Since the dataset is small and imbalanced, a transfer learning approach is applied. Moreover, as the recall of non-negative class is an important metric for this application, we propose a two-stage classification method to boost the results. The experiment results on a real dataset of TB culture test (1727 samples with 16,503 images from Tao-Yuan General Hospital, Taiwan) show that the proposed method can achieve 99% precision and 98% recall on the non-negative class. Keywords Automatic tuberculosis diagnosis · Tuberculosis culture test · Deep learning · Transfer learning · Multi-stage classification
1 Introduction Tuberculosis (TB) is an infectious disease that has a history far exceeding that of human civilization [1]. It has been one of top 10 leading causes of death which can be traced back to 1990 [2]. One influenced by TB may lead to death if not treated properly [3]. In 2017, nearly 1.6 million people died and 10 million people * Jeng‑Wei Lin [email protected] Ray‑I Chang [email protected] Yu‑Hsuan Chiu [email protected] 1
Department of Engineering Science and Ocean Engineering, National Taiwan University, Taipei, Taiwan ROC
2
Department of Information Management, Tunghai University, Box 894, No. 1727, Sec. 4, Taiwan Boulevard, Xitun District, Taichung City 407, Taiwan ROC
13
Vol.:(0123456789)
R.-I. Chang et al.
fell sick with TB [2]. To reduce the impact of TB, the World Health Organization (WHO) has set a goal to eliminate TB before 2030 [4]. Although TB is considered deadly and contagious, it can be cured and prevented by improving the detection, diagnosis, and treatment of TB. We can now cure patients better than ever with drug discovery. We can speed up the healing process and prevent infection by early and accurate detection and diagnosis. In this paper, we focus on the design of a computer-aided diagnosis (CADx) system to improve the TB diagnosis. A two-stage classification method (TSCM) that uses convolutional neural network (CNN) with transfer learning is proposed. Nowadays, there are three major ways, including acid-fast staining microscopy, chest X-rays, and culture test, to test whether a patient is infected with TB or not [5]. Recognized as the gold standard, culture test is much more accurate than other two test methods [6]. Conventionally, these test methods are performed, examined, and observed under specific environments by well-trained medical laboratory scientists. In past years, some CADx systems were designed to improve TB diagnosis process.
Data Loading...
