What question did this study set out to answer?

Reformulate volumetric 3D medical image segmentation as a sequence-to-sequence prediction task using a transformer-based encoder to overcome the receptive field limitations of standard convolutional networks.

January 1, 2022

UNETR: Transformers for 3D Medical Image Segmentation

View Full Paper

Ask AI

Bookmark

Authors

AHAli Hatamizadeh YTYucheng Tang VNVishwesh Nath

Discussion

Member takes

Overview

Deep learning evaluation demonstrates state-of-the-art 3D medical segmentation across abdominal organs and brain tumors, indicating the power of transformer encoders for spatial modeling.

Key Points

Reformulate volumetric 3D medical image segmentation as a sequence-to-sequence prediction task using a transformer-based encoder to overcome the receptive field limitations of standard convolutional networks.
Constructed UNETR, a U-shaped architecture pairing a pure transformer encoder to capture global sequence representations with a convolutional decoder connected via multiscale skip connections.
Evaluated segmentation accuracy on volumetric CT and MRI benchmarks, specifically the Multi Atlas Labeling Beyond The Cranial Vault (BTCV) multi-organ dataset and the Medical Segmentation Decathlon (MSD) brain tumor and spleen tasks.
Achieved new state-of-the-art segmentation accuracy on the BTCV multi-organ leaderboard.
Demonstrated competitive volumetric segmentation performance across multi-modality targets, including brain tumors and spleen on the MSD benchmark.

Cite This Study

Hatamizadeh et al. (2022) studied this question.

synapsesocial.com/papers/69d6e02ca0177bf533ed941c https://doi.org/10.1109/wacv51458.2022.00181

View Full Paper

Ask AI

Bookmark

January 1, 2022

UNETR: Transformers for 3D Medical Image Segmentation

View Full Paper

Ask AI

Bookmark

Authors

AHAli Hatamizadeh YTYucheng Tang VNVishwesh Nath

Discussion

Member takes

Overview

Deep learning evaluation demonstrates state-of-the-art 3D medical segmentation across abdominal organs and brain tumors, indicating the power of transformer encoders for spatial modeling.

Key Points

Reformulate volumetric 3D medical image segmentation as a sequence-to-sequence prediction task using a transformer-based encoder to overcome the receptive field limitations of standard convolutional networks.
Constructed UNETR, a U-shaped architecture pairing a pure transformer encoder to capture global sequence representations with a convolutional decoder connected via multiscale skip connections.
Evaluated segmentation accuracy on volumetric CT and MRI benchmarks, specifically the Multi Atlas Labeling Beyond The Cranial Vault (BTCV) multi-organ dataset and the Medical Segmentation Decathlon (MSD) brain tumor and spleen tasks.
Achieved new state-of-the-art segmentation accuracy on the BTCV multi-organ leaderboard.
Demonstrated competitive volumetric segmentation performance across multi-modality targets, including brain tumors and spleen on the MSD benchmark.

Cite This Study

Hatamizadeh et al. (2022) studied this question.

synapsesocial.com/papers/69d6e02ca0177bf533ed941c https://doi.org/10.1109/wacv51458.2022.00181

View Full Paper

Ask AI

Bookmark