Improvement diagnostic accuracy of sinusitis recognition in paranasal sinus X-ray using multiple deep learning models

Abstract

Background: Sinus X-ray imaging is still used in the initial evaluation of paranasal sinusitis, which is diagnosed by the opacification or air/fluid level in the sinuses and best seen in the Waters' view of the paranasal sinus (PNS). The objective of this study was to investigate the feasibility of recognizing the maxillary sinusitis features using PNS X-ray images, as well as to propose the most effective method of determining a reasonable consensus using multiple deep learning models.

Methods: A total of 4,860 patients, which included 2,430 normal and maxillary sinusitis subjects each, underwent Waters' view PNS X-ray scan. The datasets were randomly split into training (70%), validation (15%), and test (15%) subsets. We implemented a majority decision algorithm to determine a reasonable consensus using three multiple convolutional neural network (CNN) models: VGG-16, VGG-19, and ResNet-101. The performance of sinusitis detection was evaluated with quantitative accuracy (ACC) and activation maps.

Results: We compared the results of our approaches with ACC and activation maps. ACC [and area under the curve (AUC)] of the internal test dataset was evaluated as 87.4% (0.891), 90.8% (0.891), 93.7% (0.937), and 94.1% (0.948) for VGG-16, VGG-19, ResNet-101, and the majority decision, respectively. ACC (and AUC) of the external test dataset was evaluated as 87.58% (0.877), 87.58% (0.877), 92.12% (0.929), and 94.12% (0.942) for VGG-16, VGG-19, ResNet-101, and the majority decision, respectively. Majority decision algorithms can detect missing and correct lesions using a compensation function of the majority decision.

Conclusions: The majority decision algorithm showed high accuracy and significantly more accurate lesion detection compared with those of individual CNN models. The proposed deep learning method with PNS X-ray images can be used as an adjunct to classify maxillary sinusitis.

Keywords: Sinusitis; convolutional neural network (CNN); deep learning; majority decision; paranasal sinus (PNS) X-ray.

Figure 1

Representation of the pre-processed PNS X-ray images. (A) Original PNS X-ray image; (B) patch that was extracted to contain sufficient maxillary sinus region from the original image (dotted bounding square box); (C,D,F,G) rotated images with −10 (C), −30 (D), 10 (F), and 30 degree (G) rotations; (E) mirrored image that was reversed left and right for data augmentation. PNS, paranasal sinus.

Figure 2

Representation of the process steps for the majority decision algorithm.

Figure 3

Comparison of ROC curve analysis of the multiple CNN models and the proposed majority decision for the classification of maxillary sinusitis. ROC, receiver operating characteristic; CNN, convolutional neural network.

Figure 4

Performance evaluation to recognize the sinusitis features for each CNN model and the majority decision algorithm. Input patched images from patients with left maxillary sinusitis subjects (A,B), patients with right maxillary sinusitis (C,D), and patients with bilateral maxillary sinusitis (E,F). Internal test dataset (A, C, and E) and temporal test dataset subjects (B, D, and F). CNN, convolutional neural network.

Figure 5

Performance evaluation to recognize the features of normal subjects for each CNN model and the majority decision algorithm. Internal test dataset (A, B, and C) and temporal test dataset subjects (D, E, and F). CNN, convolutional neural network.