Study on Image Processing Method and Data Augmentation for Chest X-Ray Nodule Detection with YOLOv5 Algorithm
| Autor: | ADINDRA VICKAR EGA, WUWUS ARDIATNA |
|---|---|
| Jazyk: | indonéština |
| Rok vydání: | 2023 |
| Předmět: | |
| Zdroj: | Jurnal Elkomika, Vol 11, Iss 2 (2023) |
| Druh dokumentu: | article |
| ISSN: | 2338-8323 2459-9638 |
| DOI: | 10.26760/elkomika.v11i2.424 |
| Popis: | ABSTRAK Saat ini, deteksi dini kanker paru-paru dapat dilakukan dengan sistem Computer Aided Diagnosis (CAD) berbasis AI. Oleh karena itu, peningkatan perfoma sistem CAD sangat diperlukan. Dalam studi ini, berbagai teknik pengolahan citra dan augmentasi data diterapkan untuk mengevaluasi performa deteksi nodul paru-paru pada citra X-Ray dada dengan algoritma YOLOv5. Dataset publik yang terdiri dari 1500 data latih dan 516 data uji beserta dengan anotasi nodulnya digunakan dalam simulasi. Hasil simulasi menunjukkan bahwa model YOLOv5 menghasilkan presisi, recall, dan nilai mAP yang tinggi dengan nilai masing-masing 0,811, 0,776, dan 0,858, walaupun tidak menggunakan teknik pengolahan citra dan augmentasi data. Hasil validasi silang dengan dengan dataset publik JSRT dengan augmentasi data tiga kali menunjukkan bahwa YOLOv5s memiliki performa yang lebih baik untuk deteksi nodul pada paru-paru dibandingkan variasi model YOLOv5s lainnya, dengan nilai presisi 0,719 dan nilai recall 0,630. Kata Kunci: nodul paru, YOLOv5, X-Ray dada, pengolahan citra, augmentasi data  ABSTRACT Recently, early detection of lung cancer can be performed with AI-based Computer Aided Diagnosis (CAD) system. Therefore, the performance improvement of CAD is urgently needed. In this study, various image processing and data augmentation techniques were carried out to evaluate the performance of lung nodule detection on chest X-Ray images using YOLOv5 object detection algorithms. Public dataset consist of 1500 train and 516 test data along with the annotated nodules were used. The simulation results showed that the YOLOv5 model produced high precision, recall, and mAP@0.5 values of 0.811, 0.776, and 0.858, respectively although no data augmentation and image processing techniques were performed on the previous dataset. Cross-validation results with JSRT public dataset with three times the augmentation data sets showed that YOLOv5s has better performance for nodule detection with the other model with precision and recall of 0.719 and 0.630, respectively. Keywords: lung nodule, YOLOv5, X-Ray, image processing, data augmentation |
| Databáze: | Directory of Open Access Journals |
| Externí odkaz: |
|