The success of the convolutional neural network (CNN) comes with a tremendous growth of diverse CNN structures, making it hard to deploy on limited-resource platforms. These over-sized models contain a large amount of filters in the convolutional layers, which are responsible for almost 99% of the computation. The key question here arises: Do we really need all those filters? By removing entire filters, the computational cost can be significantly reduced. Hence, in this article, a filter pruning method, a process of discarding a subset of unimportant or weak filters from the original CNN model, is proposed, which alleviates the shortcomings of over-sized CNN architectures at the cost of storage space and time. The proposed filter pruning strategy is adopted to compress the model by assigning additional importance weights to convolutional filters. These additional importance weights help each filter learn its responsibility and contribute more efficiently. We adopted different initialization strategies to learn more about filters from different aspects and prune accordingly. Furthermore, unlike existing pruning approaches, the proposed method uses a predefined error tolerance level instead of the pruning rate. Extensive experiments on two widely used image segmentation datasets: Inria and AIRS, and two widely known CNN models for segmentation: TernausNet and standard U-Net, verify that our pruning approach can efficiently compress CNN models with almost negligible or no loss of accuracy. For instance, our approach could significantly reduce 85% of all floating point operations (FLOPs) from TernausNet on Inria with a negligible drop of 0.32% in validation accuracy. This compressed network is six-times smaller and almost seven-times faster (on a cluster of GPUs) than that of the original TernausNet, while the drop in the accuracy is less than 1%. Moreover, we reduced the FLOPs by 84.34% without significantly deteriorating the output performance on the AIRS dataset for TernausNet. The proposed pruning method effectively reduced the number of FLOPs and parameters of the CNN model, while almost retaining the original accuracy. The compact model can be deployed on any embedded device without any specialized hardware. We show that the performance of the pruned CNN model is very similar to that of the original unpruned CNN model. We also report numerous ablation studies to validate our approach.
Multispectral imaging devices incorporating up to 256 different spectral channels have recently become available for various healthcare applications, as e.g. laparoscopy, gastroscopy, dermatology or perfusion imaging for wound analysis. Currently, the use of such devices is limited due to very high investment costs and slow capture times. To compensate these shortcomings, single sensors with spectral masking on the pixel level have been proposed. Hence, adequate spectral reconstruction methods are needed. Within this work, two deep convolutional neural networks (DCNN) architectures for spectral image reconstruction from single sensors are compared with each other. Training of the networks is based on a huge collection of different MSI imagestacks, which have been subsampled, simulating 16-channel single sensors with spectral masking. We define a training, validation and test set (‘HITgoC’) resulting in 351 training (631.128 sub-images), 99 validation (163.272 sub-images) and 51 test images. For the application in the field of neurosurgery an additional testing set of 36 image stacks from the Nimbus data collection is used, depicting MSI brain data during open surgery. Two DCNN architectures were compared to bilinear interpolation (BI) and an intensity difference (ID) algorithm. The DCNNs (ResNet-Shinoda) were trained on HITgoC and consist of a preprocessing step using BI or ID and a refinement part using a ResNet structure. Similarity measures used were PSNR, SSIM and MSE between predicted and reference images. We calculated the similarity measures for HitgoC and Nimbus data and determined differences of the mean similarity measure values achieved with the ResNet-ID and baseline algorithms such as BI algorithm and ResNet-Shinoda. The proposed method achieved better results against BI in SSIM (.0644 vs. .0252), PSNR (15.3 dB vs. 9.1 dB) and 1-MSE*100 (.0855 vs. .0273) and compared to ResNet-Shinoda in SSIM (.0103 vs. .0074), PSNR (3.8 dB vs. 3.6 dB) and 1-MSE*100 (.0075 vs. .0047) for HITgoC/Nimbus. In this study, significantly better results for spectral reconstruction in MSI images of open neurosurgery was achieved using a combination of ID-interpolation and ResNet structure compared to standard methods.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.