Resource Optimized Neural Architecture Search for 3D Medical Image Segmentation

Bae, Woong; Lee, Seung-Ho; Lee, Yeha; Park, Beomhee; Chung, Minki; Jung, Kyung‐Ho

doi:10.1007/978-3-030-32245-8_26

Cited by 31 publications

(24 citation statements)

References 12 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…e situation is much more complicated, and the core is to solve the problem of efficiency and quality comprehensively. Efficiency, in this case, is how to use as few resources as possible (computer equipment, network bandwidth, and time) to complete a predetermined amount of web page collection [17,18]. In the occasion of bulk collection, usually considering about half a month to collect the web page is naturally the more the better.…”

Section: Architecture Design and Model Buildingmentioning

confidence: 99%

Research on Vertical Search Method of Multidimensional Resources in English Discipline Based on Edge Computing

Xie

2021

Mobile Information Systems

View full text Add to dashboard Cite

The traditional vertical search method only considers the content of the webpage, and the global master node is not enough, which will lead to premature convergence and fall into the local optimum, resulting in insufficient multi-dimensional search of resources. Therefore, this paper proposes a multidimensional resource vertical edge based on the calculation of English subject search method. This paper analyzes the architecture of search engine firstly and then introduces the multiaccess edge computing architecture. At last, it constructs the vertical search task computing model of multidimensional resources in English discipline. By associating and traversing the attributes of multidimensional resources of English discipline, the vertical search of attribute information is realized offline, and the vertical search method of multidimensional resources of English discipline based on edge calculation is designed. In order to verify the effectiveness of the proposed method, a comparative experiment is designed. Experimental results show that the method can improve the resource search ratio and recall ratio, and it can also effectively improve the search efficiency. For an English subject resource data of 50 MB, the calculation methods of edge multidimensional resource data search recall rate can reach 97% and multidimensional resource data search time consumption is only 39 ms. The experimental results show that the performance of English subject multidimensional resources vertical search is much better.

show abstract

Section: Architecture Design and Model Buildingmentioning

confidence: 99%

Research on Vertical Search Method of Multidimensional Resources in English Discipline Based on Edge Computing

Xie

2021

Mobile Information Systems

View full text Add to dashboard Cite

show abstract

“…We adopt the same image pre-processing strategy in [7]. Since the annotation of test datasets are not publicly available, we report the 5-fold cross-validation results as in [1,7,8]. We also report the validation results on the 2D lesion segmentation dataset released by the Skin Lesion Segmentation and Classification 2018 challenge [5], which provides 2594 training images.…”

Section: Datasets and Settingsmentioning

confidence: 99%

“…When we finish searching and pruning the network, we retrain the derived network from scratch. The computation of UXNet is cheap by training 1.5 days on two TitanXP GPUs for brain task, which is cheaper than RONASMIS [1] that trained 3.1 days on one RTX 2080Ti GPU and SCNAS [8] that trained one day on four V100 GPUs. Please refer to appendix for more training details of each dataset.…”

Section: Datasets and Settingsmentioning

confidence: 99%

“…The goal of NAS is to discover better neural network architectures with the higher performance, the fewer parameters, and even lower computation cost [2,13]. For medical image segmentation, Weng et al [18], Kim et al [8], Bae et al [1], Zhu et al [23] explore to search the building blocks to construct UNet [15] structure in a gradient-based manner or reinforce learning methods. Yu et al [20] further develop a more effective search strategy to alliterative the huge memory-cost problems caused by the 3D task.…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

UXNet: Searching Multi-level Feature Aggregation for 3D Medical Image Segmentation

Zhang

et al. 2020

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Aggregating multi-level feature representation plays a critical role in achieving robust volumetric medical image segmentation, which is important for the auxiliary diagnosis and treatment. Unlike the recent neural architecture search (NAS) methods that typically searched the optimal operators in each network layer, but missed a good strategy to search for feature aggregations, this paper proposes a novel NAS method for 3D medical image segmentation, named UXNet, which searches both the scale-wise feature aggregation strategies as well as the blockwise operators in the encoder-decoder network. UXNet has several appealing benefits. (1) It significantly improves flexibility of the classical UNet architecture, which only aggregates feature representations of encoder and decoder in equivalent resolution. (2) A continuous relaxation of UXNet is carefully designed, enabling its searching scheme performed in an efficient differentiable manner. (3) Extensive experiments demonstrate the effectiveness of UXNet compared with recent NAS methods for medical image segmentation. The architecture discovered by UXNet outperforms existing state-of-the-art models in terms of Dice on several public 3D medical image segmentation benchmarks, especially for the boundary locations and tiny tissues. The searching computational complexity of UXNet is cheap, enabling to search a network with best performance less than 1.5 days on two TitanXP GPUs.

show abstract

“…AutoSNAP combines the flexibility of NASNet with the speed of DARTS by introducing an intuitive yet succinct representation (instead of NAS units) and improving the efficient search and optimization strategy. The medical imaging community has recently confirmed the potential of NAS methods to segmentation [3,[16][17][18] with adaptations for scalable [6] and resource-constrained [1] environments. We are not aware of any application of NAS to CAI.…”

Section: Introductionmentioning

confidence: 98%

AutoSNAP: Automatically Learning Neural Architectures for Instrument Pose Estimation

Kügler¹,

Uecker²,

Kuijper³

et al. 2020

Preprint

View full text Add to dashboard Cite

Despite recent successes, the advances in Deep Learning have not yet been fully translated to Computer Assisted Intervention (CAI) problems such as pose estimation of surgical instruments. Currently, neural architectures for classification and segmentation tasks are adopted ignoring significant discrepancies between CAI and these tasks. We propose an automatic framework (AutoSNAP) for instrument pose estimation problems, which discovers and learns architectures for neural networks. We introduce 1) an efficient testing environment for pose estimation, 2) a powerful architecture representation based on novel Symbolic Neural Architecture Patterns (SNAPs), and 3) an optimization of the architecture using an efficient search scheme. Using AutoSNAP, we discover an improved architecture (SNAPNet) which outperforms both the hand-engineered i3PosNet and the state-of-the-art architecture search method DARTS.

show abstract

Resource Optimized Neural Architecture Search for 3D Medical Image Segmentation

Cited by 31 publications

References 12 publications

Research on Vertical Search Method of Multidimensional Resources in English Discipline Based on Edge Computing

Research on Vertical Search Method of Multidimensional Resources in English Discipline Based on Edge Computing

UXNet: Searching Multi-level Feature Aggregation for 3D Medical Image Segmentation

AutoSNAP: Automatically Learning Neural Architectures for Instrument Pose Estimation

Contact Info

Product

Resources

About