DCSLK: Combined large kernel shared convolutional model with dynamic channel Sampling

This study centers around the competition between Convolutional Neural Networks (CNNs) with large convolutional kernels and Vision Transformers in the domain of computer vision, delving deeply into the issues pertaining to parameters and computational complexity that stem from the utilization of lar...

Full description

Saved in:
Bibliographic Details
Main Authors: Zongren Li, Shuping Luo, Hongwei Li, Yanbin Li
Format: Article
Language:English
Published: Elsevier 2025-07-01
Series:NeuroImage
Subjects:
Online Access:http://www.sciencedirect.com/science/article/pii/S1053811925002836
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1849325972131676160
author Zongren Li
Shuping Luo
Hongwei Li
Yanbin Li
author_facet Zongren Li
Shuping Luo
Hongwei Li
Yanbin Li
author_sort Zongren Li
collection DOAJ
description This study centers around the competition between Convolutional Neural Networks (CNNs) with large convolutional kernels and Vision Transformers in the domain of computer vision, delving deeply into the issues pertaining to parameters and computational complexity that stem from the utilization of large convolutional kernels. Even though the size of the convolutional kernels has been extended up to 51×51, the enhancement of performance has hit a plateau, and moreover, striped convolution incurs a performance degradation. Enlightened by the hierarchical visual processing mechanism inherent in humans, this research innovatively incorporates a shared parameter mechanism for large convolutional kernels. It synergizes the expansion of the receptive field enabled by large convolutional kernels with the extraction of fine-grained features facilitated by small convolutional kernels. To address the surging number of parameters, a meticulously designed parameter sharing mechanism is employed, featuring fine-grained processing in the central region of the convolutional kernel and wide-ranging parameter sharing in the periphery. This not only curtails the parameter count and mitigates the model complexity but also sustains the model's capacity to capture extensive spatial relationships. Additionally, in light of the problems of spatial feature information loss and augmented memory access during the 1 × 1 convolutional channel compression phase, this study further puts forward a dynamic channel sampling approach, which markedly elevates the accuracy of tumor subregion segmentation. To authenticate the efficacy of the proposed methodology, a comprehensive evaluation has been conducted on three brain tumor segmentation datasets, namely BraTs2020, BraTs2024, and Medical Segmentation Decathlon Brain 2018. The experimental results evince that the proposed model surpasses the current mainstream ConvNet and Transformer architectures across all performance metrics, proffering novel research perspectives and technical stratagems for the realm of medical image segmentation.
format Article
id doaj-art-8c1f473e6e324a5ab1ef624b0119e30f
institution Kabale University
issn 1095-9572
language English
publishDate 2025-07-01
publisher Elsevier
record_format Article
series NeuroImage
spelling doaj-art-8c1f473e6e324a5ab1ef624b0119e30f2025-08-20T03:48:15ZengElsevierNeuroImage1095-95722025-07-0131512128010.1016/j.neuroimage.2025.121280DCSLK: Combined large kernel shared convolutional model with dynamic channel SamplingZongren Li0Shuping Luo1Hongwei Li2Yanbin Li3Doctoral Candidate, Information Security, Medical Image Segmentation, Xinjiang University, Urumqi, ChinaDepartment of Gastroenterology, Joint Logistics Support Force of the Chinese People's Liberation Army, Multi-Modal Medical Image Segmentation, No 940 Hospital, Lanzhou, China; Corresponding author.Information Department, Bio-information Security, Joint Logistics Support Force of the Chinese People's Liberation Army, No 940 Hospital, Lanzhou, ChinaSchool of Computer Science and Technology, Medical Image Segmentation, Xinjiang University, Urumqi, ChinaThis study centers around the competition between Convolutional Neural Networks (CNNs) with large convolutional kernels and Vision Transformers in the domain of computer vision, delving deeply into the issues pertaining to parameters and computational complexity that stem from the utilization of large convolutional kernels. Even though the size of the convolutional kernels has been extended up to 51×51, the enhancement of performance has hit a plateau, and moreover, striped convolution incurs a performance degradation. Enlightened by the hierarchical visual processing mechanism inherent in humans, this research innovatively incorporates a shared parameter mechanism for large convolutional kernels. It synergizes the expansion of the receptive field enabled by large convolutional kernels with the extraction of fine-grained features facilitated by small convolutional kernels. To address the surging number of parameters, a meticulously designed parameter sharing mechanism is employed, featuring fine-grained processing in the central region of the convolutional kernel and wide-ranging parameter sharing in the periphery. This not only curtails the parameter count and mitigates the model complexity but also sustains the model's capacity to capture extensive spatial relationships. Additionally, in light of the problems of spatial feature information loss and augmented memory access during the 1 × 1 convolutional channel compression phase, this study further puts forward a dynamic channel sampling approach, which markedly elevates the accuracy of tumor subregion segmentation. To authenticate the efficacy of the proposed methodology, a comprehensive evaluation has been conducted on three brain tumor segmentation datasets, namely BraTs2020, BraTs2024, and Medical Segmentation Decathlon Brain 2018. The experimental results evince that the proposed model surpasses the current mainstream ConvNet and Transformer architectures across all performance metrics, proffering novel research perspectives and technical stratagems for the realm of medical image segmentation.http://www.sciencedirect.com/science/article/pii/S1053811925002836Large convolutional kernelsReceptive fieldParameter sharing mechanismChannel compressionDynamic channel sampling
spellingShingle Zongren Li
Shuping Luo
Hongwei Li
Yanbin Li
DCSLK: Combined large kernel shared convolutional model with dynamic channel Sampling
NeuroImage
Large convolutional kernels
Receptive field
Parameter sharing mechanism
Channel compression
Dynamic channel sampling
title DCSLK: Combined large kernel shared convolutional model with dynamic channel Sampling
title_full DCSLK: Combined large kernel shared convolutional model with dynamic channel Sampling
title_fullStr DCSLK: Combined large kernel shared convolutional model with dynamic channel Sampling
title_full_unstemmed DCSLK: Combined large kernel shared convolutional model with dynamic channel Sampling
title_short DCSLK: Combined large kernel shared convolutional model with dynamic channel Sampling
title_sort dcslk combined large kernel shared convolutional model with dynamic channel sampling
topic Large convolutional kernels
Receptive field
Parameter sharing mechanism
Channel compression
Dynamic channel sampling
url http://www.sciencedirect.com/science/article/pii/S1053811925002836
work_keys_str_mv AT zongrenli dcslkcombinedlargekernelsharedconvolutionalmodelwithdynamicchannelsampling
AT shupingluo dcslkcombinedlargekernelsharedconvolutionalmodelwithdynamicchannelsampling
AT hongweili dcslkcombinedlargekernelsharedconvolutionalmodelwithdynamicchannelsampling
AT yanbinli dcslkcombinedlargekernelsharedconvolutionalmodelwithdynamicchannelsampling