Lung Segmentation Using Deep Learning

Sri, S. Vinitha; Kavya, S.

doi:10.38177/ajast.2021.5202

Cited by 4 publications

(4 citation statements)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Furthermore, although DeepLab with Xception backbone outperforms other models in terms of RVD, and the CAC-UNet (first place) achieves the highest AUC, our model performs significantly better in the other three metrics. In Figure 7 , we illustrate the results of tumour segmentation on the sample images and compare them with that of [ 20 , 21 , 37 , 38 , 39 , 40 , 41 , 42 , 43 , 44 , 45 , 46 ]. As shown in this figure, the mask predicted by the proposed network is extremely close to the ground truth.…”

Section: Experimental Resultsmentioning

confidence: 99%

“…As shown in Table 5 , our proposed model achieves the highest scores in state-of-the-art accuracy in gland segmentation tasks. Figure 8 shows the results of gland segmentation on the test set and compares them with [ 21 , 37 , 38 , 39 , 40 , 41 , 42 , 43 , 44 , 45 , 46 ]. As shown from this figure, compared with other leading works, our proposed network can significantly segment gland boundaries, as indicated in the red box.…”

Section: Experimental Resultsmentioning

confidence: 99%

“…Overall, our proposed model can capture more refined features and achieve state-of-the-art accuracy in tumour segmentation tasks. [37] 0.8150 0.9060 0.8611 0.6914 0.2852 0.6511 UNet (Backbone: Vgg11) [38] 0.8258 0.9187 0.8796 0.7081 0.2964 0.6829 UNet (Backbone: Vgg16) [39] 0.8323 0.9562 0.9351 0.7177 0.2445 0.8000 UNet (Backbone: Vgg19) [21] 0.7417 0.5875 0.3889 0.5990 0.4803 0.2987 UNet (Backbone: ResNet50) [40] 0.8197 0.9312 0.8981 0.7019 0.3652 0.7179 UNet (Backbone: DenseNet121) [41] 0.2183 0.5758 0.5092 0.1441 0.4825 0.3076 NestedUNet [42] 0.7609 0.7625 0.6481 0.6254 0.5561 0.4242 Unet3+ [43] 0.7467 0.6250 0.4450 0.6127 0.3977 0.3181 DeepLab (Backbone: Xception) [44] 0.6999 0.9500 0.9259 0.5517 0.1925 0.7778 DeepLab (Backbone: ResNet50) [44] 0.7964 0.6375 0.4629 0.6684 0.3829 0.3255 DeepLab (Backbone: Drn) [44] 0.7917 0.7125 0.5740 0.6605 0.3214 0.3783 DeepLab (Backbone: MobileNet) [44] 0.7943 0.8250 0.7407 0.6658 0.4206 0.5000 DCAN [45] 0.8322 0.9562 0.9351 0.7169 0.2291 0.8000 GCN [46] 0.6372 0.6625 0.5000 0.4903 0.5051 0.3414 SegNet [47] 0 To demonstrate our proposed method's generalizability and its performance in different contexts, we use the GlaS dataset to verify the network. As shown in Table 5, our proposed model achieves the highest scores in state-of-the-art accuracy in gland segmentation tasks.…”

Section: Resultsmentioning

confidence: 99%

“…UNet (Baseline) [37] 0.5132 0.4339 0.8125 0.3745 0.4959 0.9285 UNet (Backbone: Vgg11) [38] 0.7486 0.5068 0.9480 0.6195 0.6165 0.9313 UNet (Backbone: Vgg16) [39] 0.7324 0.6328 0.8265 0.6038 0.7378 0.8375 UNet (Backbone: Vgg19) [21] 0.7289 0.5979 0.8975 0.5999 0.7595 0.7928 UNet (Backbone: ResNet50) [40] 0.6511 0.5000 0.9375 0.5065 0.9228 0.9375 UNet (Backbone: DenseNet121) [41] 0.6491 0.5998 0.9263 0.5037 0.9046 0.9261 NestedUNet [42] 0.6003 0.4533 0.8500 0.4651 0.8031 0.9315 Unet3+ [43] 0.6650 0.6725 0.9450 0.5170 0.8459 0.9428 DeepLab (Backbone: Xception) [44] 0.6867 0.4735 0.8875 0.5564 0.4423 0.9342 DeepLab (Backbone: ResNet50) [44] 0.6887 0.4866 0.9125 0.5503 0.5648 0.9358 DeepLab (Backbone: Drn) [44] 0.7367 0.5306 0.9375 0.6039 0.6299 0.9375 DeepLab (Backbone: MobileNet) [44] 0.6839 0.4933 0.9250 0.5410 0.6062 0.9367 DCAN [45] 0.6415 0.6107 0.9177 0.4896 0.9459 0.9370 GCN [46] 0.5696 0.5079 0.6983 0.4220 0.9918 0.7863 SegNet [47] 0.5206 0.5533 0.8625 0.3799 0.3995 0.9445 Proposed 0.8865 0.8920 0.9823 0.7953 0.2128 0.9475 Figure 8. Segmentation results of different networks on the GlaS dataset.…”

Section: Dsc Auc Pa Ji Rvd Precisionmentioning

confidence: 99%

See 3 more Smart Citations

RGSB-UNet: Hybrid Deep Learning Framework for Tumour Segmentation in Digital Pathology Images

Zhao

Tie³

et al. 2023

Bioengineering

View full text Add to dashboard Cite

Colorectal cancer (CRC) is a prevalent gastrointestinal tumour with high incidence and mortality rates. Early screening for CRC can improve cure rates and reduce mortality. Recently, deep convolution neural network (CNN)-based pathological image diagnosis has been intensively studied to meet the challenge of time-consuming and labour-intense manual analysis of high-resolution whole slide images (WSIs). Despite the achievements made, deep CNN-based methods still suffer from some limitations, and the fundamental problem is that they cannot capture global features. To address this issue, we propose a hybrid deep learning framework (RGSB-UNet) for automatic tumour segmentation in WSIs. The framework adopts a UNet architecture that consists of the newly-designed residual ghost block with switchable normalization (RGS) and the bottleneck transformer (BoT) for downsampling to extract refined features, and the transposed convolution and 1 × 1 convolution with ReLU for upsampling to restore the feature map resolution to that of the original image. The proposed framework combines the advantages of the spatial-local correlation of CNNs and the long-distance feature dependencies of BoT, ensuring its capacity of extracting more refined features and robustness to varying batch sizes. Additionally, we consider a class-wise dice loss (CDL) function to train the segmentation network. The proposed network achieves state-of-the-art segmentation performance under small batch sizes. Experimental results on DigestPath2019 and GlaS datasets demonstrate that our proposed model produces superior evaluation scores and state-of-the-art segmentation results.

show abstract