18 March 2025 in IEEE Transactions on Emerging Topics in Computational Intelligence
In recent years, supervised learning using convolutional neural networks (CNN) has served as a benchmark for various medical image segmentation and classification. However, supervised learning deeply relies on large-scale annotated data, which is expensive, time-consuming, and even impractical to acquire in medical imaging applications. Moreover, effective utilization of annotation resources might not always be feasible during the annotation process. To optimize the utilization of annotation resources, a proposed active learning framework is introduced that is applicable to both 2D and 3D segmentation and classification tasks. This framework aims to reduce annotation costs by selecting more valuable samples for annotation from the pool of unlabeled data. Based on the perturbation consistency, we apply different perturbations to the input data and propose a perturbation consistency evaluation module to evaluate the consistency among predictions when applying different perturbations to the data. Subsequently, we rank the consistency of each data and select samples with lower consistency as high-value candidates. These selected samples are prioritized for annotation. We extensively validated our proposed framework on three publicly available and challenging medical image datasets, Kvasir Dataset, COVID-19 Infection Segmentation Dataset, and BraTS2019 Dataset. The experimental results demonstrate that our proposed framework can achieve significantly improved performance with fewer annotations in 2D classification and segmentation and 3D segmentation tasks. The proposed framework enables more efficient utilization of annotation resources by annotating more representative samples, thus enhancing the model's robustness with fewer annotation costs.
9 October 2020 in IEEE Access
Stereo disparity estimation is a difficult and crucial task in computer vision. Although many experimental techniques have been proposed in recent years with the flourishing of deep learning, very few studies take into account the optimization of computational complexity and memory consumption. Most previous works take advantage of stacked 3D convolutional block to generate fine disparity, but with a high computational cost and a large memory consumption. Considering the aforementioned problem, in this paper, we proposed an efficient convolutional neural architecture for stereo disparity estimation. In particular, a compact and efficient multi-scale extractor named MCliqueNet with stacked CliqueBlock was proposed to extract the more refined features for constructing multi-scale cost volume. In order to reduce the computational cost and maintain the accuracy of disparity, we utilized knowledge distillation scheme to transfer contextual features from a teacher network to a student network. Furthermore, we present a novel adaptive SmoothL1 (ASL) Loss for calculating the similarity between the contextual features of the teacher network and those of the student network, resulting in a more robust distillation process. Experimental results have shown that our method achieves competitive performance on the challenging Scene Flow and KITTI benchmarks while maintaining a very fast running speed.
15 June 2024 in Expert Systems with Applications