Strided U-Net Model: Retinal Vessels Segmentation using Dice Loss

Soomro, Toufique Ahmed; Hellwich, Olaf; Afifi, Ahmed J.; Paul, Manoranjan; Gao, Junbin; Zheng, Lihong

doi:10.1109/dicta.2018.8615770

Cited by 70 publications

(42 citation statements)

References 33 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Another method proposed by Hajabdollahi et al [11], and achieved excellent performances with an AUC of 0.97 and an accuracy of 0.961 on the STARE database. Soomro et al [12] verified their method on the two benchmarking databases, and the performance of the method is highly comparable to that of other existing approaches. Tan et al [13] suggested a 7-layer CNN based method to segmenting the retinal blood vessels.…”

Section: Literature Reviewmentioning

confidence: 86%

Mathematical Morphology based Retinal Image Blood Vessels Segmentation

2019

IJITEE

View full text Add to dashboard Cite

It is necessary to verify the state of blood vessel network in the retina for diagnosing various issues associated with eyes. In this research paper, an involuntary retinal vessel segmentation using mathematical morphology is proposed. The contrast of the retinal images is enhanced by contrast limited adaptive histogram equalisation technique. Ten blood vessels of the enhanced retinal image are detected using morphological processing. The hysteresis thresholding is applied on the blood vessels detected image to remove the unwanted back ground detail. Finally the properly segmented binary image of the retinal vessel is obtained using post processing process. Results of the presented method are verified by using most widely used for benchmarking retinal image databases such as, Child Heart and Health Study in England (CHASE_DB1) and Digital Retinal Images for Vessel Extraction (DRIVE) database by computing the evaluation metrics such as sensitivity, specificity, accuracy and precision. The better evaluation metrics achieved for the DRIVE dataset are 0.7493, 0.9687, 0.9524 and 0.6590, and the worst values are 0.6621, 0.9411, 0.9137 and 0.5491. The best evaluation metrics values for the CHASE_DB1 dataset are 0.5058, 0.8947, 0.9382 and 0.8856, and the worst values are 0.5639, 0.9581, 0.9137 and 0.7110. The investigational results show that the suggested approach provides the excellent accuracy in comparison with other approaches

show abstract

Section: Literature Reviewmentioning

confidence: 86%

Mathematical Morphology based Retinal Image Blood Vessels Segmentation

2019

IJITEE

View full text Add to dashboard Cite

show abstract

“…The inputs of our models are images and ground truths. We use the BCELoss [ 41 ] and Dice loss [ 42 , 43 ] as our loss function. The BCELoss is designed for binary classification.…”

Section: Proposed Methodsmentioning

confidence: 99%

UnetDVH-Linear: Linear Feature Segmentation by Dilated Convolution with Vertical and Horizontal Kernels

Liao

Cao

et al. 2020

Sensors

View full text Add to dashboard Cite

Linear feature extraction is crucial for special objects in semantic segmentation networks, such as slot marking and lanes. The objects with linear characteristics have global contextual information dependency. It is very difficult to capture the complete information of these objects in semantic segmentation tasks. To improve the linear feature extraction ability of the semantic segmentation network, we propose introducing the dilated convolution with vertical and horizontal kernels (DVH) into the task of feature extraction in semantic segmentation networks. Meanwhile, we figure out the outcome if we put the different vertical and horizontal kernels on different places in the semantic segmentation networks. Our networks are trained on the basis of the SS dataset, the TuSimple lane dataset and the Massachusetts Roads dataset. These datasets consist of slot marking, lanes, and road images. The research results show that our method improves the accuracy of the slot marking segmentation of the SS dataset by 2%. Compared with other state-of-the-art methods, our UnetDVH-Linear (v1) obtains better accuracy on the TuSimple Benchmark Lane Detection Challenge with a value of 97.53%. To prove the generalization of our models, road segmentation experiments were performed on aerial images. Without data argumentation, the segmentation accuracy of our model on the Massachusetts roads dataset is 95.3%. Moreover, our models perform better than other models when training with the same loss function and experimental settings. The experiment result shows that the dilated convolution with vertical and horizontal kernels will enhance the neural network on linear feature extraction.

show abstract

“…and PCNN (Pulse-coupled neural network) [116]. CNN uses information from shallow to deep layers to determine the fine details and overall structure of retinal vessels [101][102][103][104]133]. An example of CNN network design might be an input convolution layer containing 1x28x28 patches.…”

Section: H Supervised Learning Algorithms and Deep Learningmentioning

confidence: 99%

“…Due to the consistency of the tables, the Table XIII below does not contain these parameters AUC in articles Soomro et al [101], Chudzik et al [103], Hajabdollahi et al [102], Guo et al [104], Sengür et al [110], Feng et al [112], Soomro et al [114] and AU ROC parameter in articles Mo et al [108], Lahiri et al [109], Wang et al [132], Wu et al [125] and PPV in article Feng et al [112] and visual comparison in article Gu et al [133]. It contains only Acc, Se and Sp.…”

Section: H Supervised Learning Algorithms and Deep Learningmentioning

confidence: 99%