Application of Principal Component Analysis and Adaptive Median Filter to Improve Real-Time Prostate Capsula Detection

Abstract

This study investigated the potential for using Principal Component Analysis (PCA) and Adaptive Median Filter (AMF) to improve real-time prostate capsula detection with the traditional Region-based Fully Convolutional Network (R-FCN), Faster Region-based Convolutional Neural Network (Faster R-CNN), You Only Look Once-Version 3 (YOLOv3) and Single Shot Multibox Detector (SSD) algorithms. The processing steps included data augmentation (rotation, vertical flip, and horizontal flip) to increase the size of the dataset from 149 to 596 images, PCA-based feature extraction, AMF-based image denoising and a training phase incorporating the sample image set. The data were then used to test a series of combined methods that were applied to the detection of prostate capsula (PC). The results showed that application of PCA and AMF to Faster R-CNN increased the mean average precision (mAP) for the PC images by 9.4%. The application of PCA and AMF to R-FCN, YOLOv3 and SSD increased the mAP by 7.22%, 7.14% and 3.29% for the same dataset, respectively. This study represents the first application of PCA and AMF to traditional object detection algorithms, such as R-FCN, Faster R-CNN, YOLOv3, or SSD, and the improved mAP results suggest that this approach is a robust tool for improving multiple network architectures.