Papers › Medical Image Segmentation via Cascaded Attention Decoding

Medical Image Segmentation via Cascaded Attention Decoding

3 Jan 2023Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) 2023 1archive 2025-07-28

Md Mostafijur Rahman, Radu Marculescu

Transformers have shown great promise in medical image segmentation due to their ability to capture long-range dependencies through self-attention. However, they lack the ability to learn the local (contextual) relations among pixels. Previous works try to overcome this problem by embedding convolutional layers either in the encoder or decoder modules of transformers thus ending up sometimes with inconsistent features. To address this issue, we propose a novel attention-based decoder, namely CASCaded Attention DEcoder (CASCADE), which leverages the multiscale features of hierarchical vision transformers. CASCADE consists of i) an attention gate which fuses features with skip connections and ii) a convolutional attention module that enhances the long-range and local context by suppressing background information. We use a multi-stage feature and loss aggregation framework due to their faster convergence and better performance. Our experiments demonstrate that transformers with CASCADE significantly outperform state-of-the-art CNN- and transformer-based approaches, obtaining up to 5.07% and 6.16% improvements in DICE and mIoU scores, respectively. CASCADE opens new ways of designing better attention-based decoders.

Paper PDF Code

Code

SLDGroup/CASCADE pytorch report

Repository list and official/mentioned flags are the archive's, frozen 2025-07-28. Reachability, where shown, is from one Syntology probe window (2026-09-16 to 2026-09-18); repositories not probed show nothing. GitHub stars are not tracked.

Code Syntology ran Syntology

Not run by Syntology. Nothing on this page verifies that the listed code works.

Tasks

Decoder Image Segmentation Medical Image Segmentation Polyp Segmentation Semantic Segmentation

Results from the paper archive 2025-07-28

Task	Dataset	Model	Metric	Value	Rank at snapshot	Leaderboard	Report
Medical Image Segmentation	Automatic Cardiac Diagnosis Challenge (ACDC)	TransCASCADE	Avg DSC	91.63	#12 of 20	Archive leaderboard	report
Medical Image Segmentation	Automatic Cardiac Diagnosis Challenge (ACDC)	PVT-CASCADE	Avg DSC	91.46	#13 of 20	Archive leaderboard	report
Medical Image Segmentation	CVC-ClinicDB	PVT-CASCADE	mIoU	0.8998	#16 of 48	Archive leaderboard	report
Medical Image Segmentation	CVC-ClinicDB	PVT-CASCADE	mean Dice	0.9434	#16 of 48	Archive leaderboard	report
Medical Image Segmentation	CVC-ColonDB	PVT-CASCADE	mIoU	0.7453	#10 of 25	Archive leaderboard	report
Medical Image Segmentation	CVC-ColonDB	PVT-CASCADE	mean Dice	0.8254	#10 of 25	Archive leaderboard	report
Medical Image Segmentation	ETIS-LARIBPOLYPDB	PVT-CASCADE	mIoU	0.7258	#9 of 25	Archive leaderboard	report
Medical Image Segmentation	ETIS-LARIBPOLYPDB	PVT-CASCADE	mean Dice	0.8007	#9 of 25	Archive leaderboard	report
Medical Image Segmentation	Kvasir-SEG	PVT-CASCADE	mIoU	0.8776	#21 of 58	Archive leaderboard	report
Medical Image Segmentation	Kvasir-SEG	PVT-CASCADE	mean Dice	0.9258	#21 of 58	Archive leaderboard	report
Medical Image Segmentation	MICCAI 2015 Multi-Atlas Abdomen Labeling Challenge	TransCASCADE	Avg DSC	82.68	#6 of 8	Archive leaderboard	report
Medical Image Segmentation	MICCAI 2015 Multi-Atlas Abdomen Labeling Challenge	TransCASCADE	Avg HD	17.34	#6 of 8	Archive leaderboard	report
Medical Image Segmentation	MICCAI 2015 Multi-Atlas Abdomen Labeling Challenge	PVT-CASCADE	Avg DSC	81.06	#8 of 8	Archive leaderboard	report
Medical Image Segmentation	MICCAI 2015 Multi-Atlas Abdomen Labeling Challenge	PVT-CASCADE	Avg HD	20.23	#8 of 8	Archive leaderboard	report
Polyp Segmentation	Kvasir-SEG	PVT-CASCADE	mDice	0.9258	#1 of 8	Archive leaderboard	report
Polyp Segmentation	Kvasir-SEG	PVT-CASCADE	mIoU	0.8776	#1 of 8	Archive leaderboard	report

Ranks are positions in the archive's leaderboards as they stood at the 2025-07-28 snapshot. Results published since then are not among these rows, so a rank here is not a current standing.

Methods

Absolute Position Encodings Adam Attention BPE Dense Connections Dropout Label Smoothing Layer Normalization Linear Layer Multi-Head Attention Position-Wise Feed-Forward Layer Residual Connection Softmax Transformer

Report a problem or propose a change · a person checks every report against the paper or source before anything changes; decisions are listed on /corrections