TransNetR: Transformer-based Residual Network for Polyp Segmentation with Multi-Center Out-of-Distribution Testing (MIDL 2023)
Paper Link: https://arxiv.org/pdf/2303.07428.pdf
TransNetR is an encoder decoder network which can be used for efficient biomedical image segmentation for both in-distribution and out-of-distribution datasets.
Figure 1: Illustration of different scenarios expected to arise in real-world settings. The proposed work conducted both in-distribution and out-of-distribution validation process. C1 to C6 represent the different centers data present in PolypGen dataset width=50% height=50%
Figure 2: Block diagram of TransNetR along with the Residual Transformer block
Figure 3: Qualitative example showing polyp segmentation on Kvasir-SEG
Table 1: Quantitative results on the Kvasir-SEG test dataset. The parameters are in Mil- lions and Flops are in GMac.
Figure 4: Cross-data result when models trained on Kvasir-SEG & tested on BKAI-IGH.LeakyReLU activation function. Finally, the output from the LeakyReLU is passed througha residual block which acts as the output of the residual transformer block.
Figure 5: Center-wise example images from the PolypGen dataset. Here, the variabilityamong the dataset from different centers can be observed. There is a differencein image resolutions and sizes, shapes, colors, textures and appearances and col-lection protocols.Figure 6: Qualitative result when the TransNetR is trained on Kvasir-SEG and tested on(a) PolypGen (center 6 (C6)) and (b) PolypGen (center 1 (C1)).13
Figure 6: Qualitative result when the TransNetR is trained on Kvasir-SEG and tested on(a) PolypGen (center 6 (C6))
Figure 7: Qualitative result when the TransNetR is trained on Kvasir-SEG and tested on PolypGen (center 1 (C1))
Please cite our paper if you find the work useful:
@INPROCEEDINGS{JhaTrans2023,
author={D.{Jha} and N.{Tomar} and V.{Sharma} and U.{Bagci}},
booktitle={Proceedings of the Medical Imaging with Deep Learning},
title={TransNetR: Transformer-based Residual Network for Polyp Segmentation with Multi-Center Out-of-Distribution Testing},
year={2023}}
Please contact [email protected] and [email protected] for any further questions.