结 合 PCANet与线性判别分析的视网膜光学相干断层扫描图像分类

Objective Optical coherence tomography (OCT) is a 3D scanning imaging technology that has been widely used in ophthalmology as a clinical auxiliary to identify various eye lesions. Therefore, the classification technique of retinal O C T images is greatly important for the detection and treatment of retinopathy. Many effective O C T classification algorithms have been recently developed, and almost all these have artificial design features ； however, retinal O C T images acquired from clinic usually contains a complex pathological structure. Therefore, the features from O C T images must be directly learned. Principal component analysis network (PCANet) is a simple version of convolutional neural network, which can directly extract the texture features of images, whereas features extracted by linear discriminant analysis (L D A) are more distinguishable for image classification. Combining the advantages of these two methods, this paper presents a PCANet with L D A (PCANet-LDA) for the automatic classification of three types of retinal O C T images, including age-related macular degeneration (A M D), diabetic macular edema (D M E), and normal (NOR). Method The proposed PCANet-LDA algorithm adds an L D A supervisory layer based on the PCANet to allow the supervision of extracted image features by class labels. This algorithm can be implemented in three steps. The first step is the O C T image preprocessing, which involves a series of preprocessing including perceiving, fitting, and normalizing stages on retinal O C T images to obtain an interested retinal region for image classification. The second step is the PCANet feature extraction, where the preprocessed O C T images are sent into a P C A convolution layer with two stages and a nonlinear output layer. In the P C A convolution layer, P C A filter banks are learned, and the P C A features of retinal O C T images can be extracted. In the nonlinear output layer, the extracted P C A features are translated to PCANet features of the input images by some basic data-processing components, including binary hashing and blockwise histograms. The third step is the L D A supervisory layer, which uses the L D A idea to learn an L D A matrix from the PCANet features with class labels of A M D, D M E, and NOR. Then, the L D A matrix is used to project PCANet features into a low-dimensional space to make the projected features more distinguishable for classification. Finally, the projected features are used to train a linear support vector machine and classify the retinal O C T images. Result Both experiments are done on two retinal O C T dataset, including the clinic dataset obtained from a hospital and Duke dataset. First, the comparative examples of A M D, D M E and N O R retinal O C T images before and after preprocessing shows that the image preprocessing cuts out the non-retinal regions in the O C T image, leaving the meaningful retinal areas. Moreover, the remaining retina is rotated to a unified horizontal state to reduce the impact of inconsistent direction of retina on classification. Then, the sample PCANet feature maps extracted from A M D and D M E retinal O C T images show that the P C A filter trained by PCANet tends to capture meaningful pathological structure information, which contributes to the classification of retinal O C T images. Finally, the correct classification rates of the PCANet algorithm, the ScSPM algorithm, and the PCANet-LDA algorithm proposed in this paper are compared. On the clinic dataset, the overall correct classification rate of the PCANet-LDA algorithm is 97.2 0 %, which is 3.7 7 % higher than that of the PCANet algorithm and slightly higher than that of the ScSPM algorithm. O n the Duke dataset, the overall correct classification rate of the PCANet-LDA algorithm is 99.5 2 %, which is 1.6 4 % higher than that of the PCANet algorithm and a slightly higher than that of the ScSPM algorithm. Conclusion The PCANet algorithm can extract effective features. Accordingly, the PCANet-LDA algorithm obtains more distinguishing features by L D A method, to yield a higher correct classification rate than that of the PCANet and ScSPM algorithms ； the latter is a state-of-the-art two-dimensional O C T image classification of the retina. Therefore, the proposed PCANet-LDA algorithm is effective, advanced in the classification of retinal O C T images, and can be a baseline algorithm for retinal O C T image classification.