Optimasi Convolutional Neural Network Untuk Deteksi Covid-19 pada X-ray Thorax Berbasis Dropout

Penulis

I Gede Totok Suryawan, I Putu Agus Eka Darma Udayana

Abstrak

Pandemi COVID-19 yang melanda Indonesia sejak pertengahan tahun 2020 telah memberikan dampak luar biasa pada infrastruktur medis di Indonesia. Angka rata-rata penyebaran virus COVID-19 yang cukup tinggi membuat monitoring bed occupancy rate menjadi sebuah tantangan tersendiri. Dengan adanya penetrasi Artificial Intelligence yang tepat pada sistem medis di Indonesia, diharapkan dapat membantu terjadinya transfer knowledge antar paramedis menjadi lebih efektif. Salah satunya dengan menggunakan Deep learning yaitu Convolutional Neural Network (CNN) yang sudah terbukti merupakan salah satu metode yang dapat digunakan untuk melakukan skrining pasien dan mendeteksi COVID-19. Namun untuk melatih sebuah classifier CNN yang ampuh dan siap digunakan di dunia nyata membutuhkan computing power yang besar dan umumnya training rate yang lama.  Penelitian ini bertujuan untuk membuat arsitektur jaringan syaraf tiruan berbasis deep learning yang lebih cepat dan efisien dengan pembuatan network yang  lebih ramping sehingga lebih mudah dibuat oleh orang lain tanpa harus memiliki computing power yang besar. Metode yang digunakan adalah dengan menyisipkan dropout layer pada sistem jaringan syaraf tiruan. Metode ini akan memaksa sistem untuk belajar memakai rute yang tersingkat dengan cara menghilangkan beberapa node secara acak. Arsitektur ini kemudian diuji pada data ronsen thorax penyintas COVID-19 dan kemudian dibandingkan dengan arsitektur lainnya yang sama-sama memakai pendekatan deep learning. Setelah ditraning menggunakan 500 data COVID-19 thorax X-Ray public database dan diuji dengan jumlah data yang sama, classifier yang menggunakan arsitektur ini mampu menghasilkan akurasi sebesar 95,20%, precision 94,80%, recall 95,58%, specificity 94,88%, NVP sebesar 95,60%, F-Score sebesar 95,18 dan dapat menghemat waktu training sampai 62% dibandingkan dengan arsitektur deep learning lainnya.

 

Abstract

The COVID-19 pandemic that hit Indonesia in mid-2020 had a tremendous impact on medical infrastructure in Indonesia. The virus made monitoring the bed occupancy rate became a challenge in itself. New approach can be taken to fight the crisis. The Convolutional Neural Network (CNN), which has proved to be one of the methods that can use to screen patients and detect COVID-19.also have its own problem because it requires enormous computing power and generally a long training rate. Therefore, this study aimed to tackle that problem by creating a leaner network. Thus, it is easier for others to build without having enormous computing power. The method used was to insert a dropout layer on the artificial network system. This method will force the system to learn using the shortest route by eliminating some nodes at random. Then, this architecture was tested on chest X-ray data of COVID-19 survivors and compared with other architectures that both used a deep learning approach. It proved that when this system was tested with COVID-19 thorax x-ray public database data, the classifier that used this architecture could achieve an accuracy rate of 95.20% followed by precision and recall value reaching 94.80% and 94.80%. respectively and last but not least F-score of 95.18% and Negative Predictive value of 95.60%  It could also save training time up to 62% compared to other deep learning architectures. Using dropout layers proved could produce more efficient layers and more powerful classifiers while keeping training time to a minimum.

Teks Lengkap:

PDF

Referensi


ABDAR, M., dkk. 2021.UncertaintyFuseNet: Robust Uncertainty-aware Hierarchical Feature Fusion with Ensemble Monte Carlo Dropout for COVID-19 Detection. IEEE TRANSACTIONS, pp. 1–16. Available at: http://arxiv.org/abs/2105.08590.

DIAZ-ESCOBAR, J., dkk. 2021. Deep-learning based detection of COVID-19 using lung ultrasound imagery. PLoS ONE, 16(8 August), pp. 1–21. doi: 10.1371/journal.pone.0255886.

ELGENDI, M., dkk. 2021. The Effectiveness of Image Augmentation in Deep Learning Networks for Detecting COVID-19: A Geometric Transformation Perspective. Frontiers in Medicine. doi: 10.3389/fmed.2021.629134.

GHAZI, M. M. & EKENEL, H. K. 2016. A Comprehensive Analysis of Deep Learning Based Representation for Face Recognition. IEEE Computer Society Conference on Computer Vision and Pattern Recognition Workshops, pp. 102–109. doi: 10.1109/CVPRW.2016.20.

GREWAL, M. , dkk. 2018. RADnet: Radiologist level accuracy using deep learning for hemorrhage detection in CT scans. Proceedings - International Symposium on Biomedical Imaging, 2018-April(Isbi), pp. 281–284. doi: 10.1109/ISBI.2018.8363574.

GUO, J. & LI, B. 2018. The Application of Medical Artificial Intelligence Technology in Rural Areas of Developing Countries. Health Equity, 2(1), pp. 174–181. doi: 10.1089/heq.2018.0037.

HOU, T., dkk. 2020. Self-efficacy and fatigue among health care workers during COVID-19 outbreak: A moderated mediation model of posttraumatic stress disorder symptoms and negative coping. pp. 1–21. doi: 10.21203/rs.3.rs-23066/v1.

LIU, C. & ALCANTARA, M. F. 2016. Tx-Cnn : Detecting Tuberculosis In Chest X-Ray Images Using Convolutional Neural Network Partners In Health Per ´ U’. 2017 IEEE International Conference on Image Processing (ICIP), doi: 10.1109/ICIP.2017.8296695.

LIU, J., dkk. 2018. Applications of deep learning to MRI Images: A survey. Big Data Mining and Analytics, 1(1), pp. 1–18. doi: 10.26599/BDMA.2018.9020001.

NOWLAN, S. J. 2018. Simplifying Neural Networks by Soft Weight Sharing. The Mathematics of Generalization, 493, pp. 373–394. doi: 10.1201/9780429492525-13.

RAHMAN, T., dkk. 2020. Transfer learning with deep Convolutional Neural Network (CNN) for pneumonia detection using chest X-ray’, Applied Sciences (Switzerland), 10(9). doi: 10.3390/app10093233.

RICE, L., WONG, E. & KOLTER, J. Z. 2020. Overfitting in adversarially robust deep learning’, arXiv.

SHAO, S., dkk. 2020. Hardware for Machine Learning Course Overview Instructor Teaching Assistants Lectures and O ice Hours’, Berkeley University, pp. 1–5.

SHINDE, S. V & MANE, D. T. 2021. Deep Learning for COVID-19: COVID-19 Detection Based on Chest X-Ray Images by the Fusion of Deep Learning and Machine Learning Techniques’, Understanding COVID-19: The Role of Computational Intelligence. Studies in Computational Intelligence, 963, pp. 471–500. doi: 10.1007/978-3-030-74761-9_21.

SRIVASTAVA, N., dkk. 2014. Dropout: A Simple Way to Prevent Neural Networks from Overfitting. Journal of Machine Learning Research, 299(3–4), pp. 345–350. doi: 10.1016/0370-2693(93)90272-J.

SUN, G, dkk. 2018. Combined Deep Learning and Multiscale Segmentation for Rapid High Resolution Damage Mapping.

Proceedings - 2017 IEEE International Conference on Internet of Things, IEEE Green Computing and Communications, IEEE Cyber, Physical and Social Computing, IEEE Smart Data, iThings-GreenCom-CPSCom-SmartData 2017, 2018-Janua, pp. 1101–1105. doi: 10.1109/iThings-GreenCom-CPSCom-SmartData.2017.238.

SURYAWAN, I. G. T. & UDAYANA, I. P. A. E. D. 2020. A Deep Learning Approach For COVID 19 Detection Via X-Ray Image With Image Correction Method. International Journal of Engineering and Emerging Technology, 5(2), pp. 1–5.

Available at:https://ocs.unud.ac.id/index.php/ijeet/article/view/64535/37517.

doi:https://doi.org/10.24843/IJEET.2020.v05.i02.p018.

USCHER-PINES, L.,dkk. 2006. Priority setting for pandemic influenza: An analysis of national preparedness plans. PLoS Medicine, 3(10), pp. 1721–1727. doi: 10.1371/journal.pmed.0030436.

VAHRUN. 2021. Dirjen Dikti Apresiasi GeNose C19 Hasil Inovasi Karya Anak Bangsa, lldikti1.ristekdikti.go.id. Available at: https://lldikti1.ristekdikti.go.id/details/apps/2609.

ZHANG, C., dkk. (2018. A study on overfitting in deep reinforcement learning’, arXiv, pp. 1–25.




DOI: http://dx.doi.org/10.25126/jtiik.2022935143