3 February 2020 in IEEE Access
Medical image classification plays an important role in disease diagnosis since it can provide important reference information for doctors. The supervised convolutional neural networks (CNNs) such as DenseNet provide the versatile and effective method for medical image classification tasks, but they require large amounts of data with labels and involve complex and time-consuming training process. The unsupervised CNNs such as principal component analysis network (PCANet) need no labels for training but cannot provide desirable classification accuracy. To realize the accurate medical image classification in the case of a small training dataset, we have proposed a light-weighted hybrid neural network which consists of a modified PCANet cascaded with a simplified DenseNet. The modified PCANet has two stages, in which the network produces the effective feature maps at each stage by convoluting inputs with various learned kernels. The following simplified DenseNet with a small number of weights will take all feature maps produced by the PCANet as inputs and employ the dense shortcut connections to realize accurate medical image classification. To appreciate the performance of the proposed method, some experiments have been done on mammography and osteosarcoma histology images. Experimental results show that the proposed hybrid neural network is easy to train and it outperforms such popular CNN models as PCANet, ResNet and DenseNet in terms of classification accuracy, sensitivity and specificity.
1 June 2021 in Signal Processing
11 February 2023 in CAAI Transactions on Intelligence Technology
Multi‐modal brain image registration has been widely applied to functional localisation, neurosurgery and computational anatomy. The existing registration methods based on the dense deformation fields involve too many parameters, which is not conducive to the exploration of correct spatial correspondence between the float and reference images. Meanwhile, the unidirectional registration may involve the deformation folding, which will result in the change of topology during registration. To address these issues, this work has presented an unsupervised image registration method using the free form deformation (FFD) and the symmetry constraint‐based generative adversarial networks (FSGAN). The FSGAN utilises the principle component analysis network‐based structural representations of the reference and float images as the inputs and uses the generator to learn the FFD model parameters, thereby producing two deformation fields. Meanwhile, the FSGAN uses two discriminators to decide whether the bilateral registration have been realised simultaneously. Besides, the symmetry constraint is utilised to construct the loss function, thereby avoiding the deformation folding. Experiments on BrainWeb, high grade gliomas, IXI and LPBA40 show that compared with state‐of‐the‐art methods, the FSGAN provides superior performance in terms of visual comparisons and such quantitative indexes as dice value, target registration error and computational efficiency.