FocusNet++: Attentive Aggregated Transformations for Efficient and Accurate Medical Image Segmentation

Chaitanya Kaul; N. E. Pears; Hang Dai; Roderick Murray-Smith; Suresh Manandhar

doi:10.1109/ISBI48211.2021.9433918

FocusNet++: Attentive Aggregated Transformations for Efficient and Accurate Medical Image Segmentation

Chaitanya Kaul, N. E. Pears, Hang Dai, Roderick Murray-Smith, Suresh Manandhar

Computer Science

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

Abstract

We propose a new residual block for convolutional neural networks and demonstrate its state-of-the-art performance in medical image segmentation. We combine attention mechanisms with group convolutions to create our group attention mechanism, which forms the fundamental building block of our network, FocusNet++. We employ a hybrid loss based on balanced cross entropy, Tversky loss and the adaptive logarithmic loss to enhance the performance along with fast convergence. Our results show that FocusNet++ achieves state-of-the-art results across various benchmark metrics for the ISIC 2018 melanoma segmentation and the cell nuclei segmentation datasets with fewer parameters and FLOPs.

Original language	English
Title of host publication	2021 IEEE 18th International Symposium on Biomedical Imaging (ISBI)
DOIs	https://doi.org/10.1109/ISBI48211.2021.9433918
Publication status	Published - 25 May 2021
Event	IEEE International Symposium on Biomedical Imaging, 2021: ISBI 2021 - Online, Nice, France Duration: 13 Apr 2021 → 16 Apr 2021 https://biomedicalimaging.org/2021/

Conference

Conference	IEEE International Symposium on Biomedical Imaging, 2021
Abbreviated title	ISBI 2021
Country/Territory	France
City	Nice
Period	13/04/21 → 16/04/21
Internet address	https://biomedicalimaging.org/2021/

Keywords

Group Ateention, Medical Image Segmentation, Residual Learning

Access to Document

10.1109/ISBI48211.2021.9433918

focusnetPlusPlus-AcceptedAccepted author manuscript, 608 KB

Cite this

@inproceedings{81eea53efce14a97977d22069b5a2165,

title = "FocusNet++: Attentive Aggregated Transformations for Efficient and Accurate Medical Image Segmentation",

abstract = "We propose a new residual block for convolutional neural networks and demonstrate its state-of-the-art performance in medical image segmentation. We combine attention mechanisms with group convolutions to create our group attention mechanism, which forms the fundamental building block of our network, FocusNet++. We employ a hybrid loss based on balanced cross entropy, Tversky loss and the adaptive logarithmic loss to enhance the performance along with fast convergence. Our results show that FocusNet++ achieves state-of-the-art results across various benchmark metrics for the ISIC 2018 melanoma segmentation and the cell nuclei segmentation datasets with fewer parameters and FLOPs.",

keywords = "Group Ateention, Medical Image Segmentation, Residual Learning",

author = "Chaitanya Kaul and Pears, {N. E.} and Hang Dai and Roderick Murray-Smith and Suresh Manandhar",

year = "2021",

month = may,

day = "25",

doi = "10.1109/ISBI48211.2021.9433918",

language = "English",

booktitle = "2021 IEEE 18th International Symposium on Biomedical Imaging (ISBI)",

note = "IEEE International Symposium on Biomedical Imaging, 2021 : ISBI 2021, ISBI 2021 ; Conference date: 13-04-2021 Through 16-04-2021",

url = "https://biomedicalimaging.org/2021/",

}

TY - GEN

T1 - FocusNet++

T2 - IEEE International Symposium on Biomedical Imaging, 2021

AU - Kaul, Chaitanya

AU - Pears, N. E.

AU - Dai, Hang

AU - Murray-Smith, Roderick

AU - Manandhar, Suresh

PY - 2021/5/25

Y1 - 2021/5/25

N2 - We propose a new residual block for convolutional neural networks and demonstrate its state-of-the-art performance in medical image segmentation. We combine attention mechanisms with group convolutions to create our group attention mechanism, which forms the fundamental building block of our network, FocusNet++. We employ a hybrid loss based on balanced cross entropy, Tversky loss and the adaptive logarithmic loss to enhance the performance along with fast convergence. Our results show that FocusNet++ achieves state-of-the-art results across various benchmark metrics for the ISIC 2018 melanoma segmentation and the cell nuclei segmentation datasets with fewer parameters and FLOPs.

AB - We propose a new residual block for convolutional neural networks and demonstrate its state-of-the-art performance in medical image segmentation. We combine attention mechanisms with group convolutions to create our group attention mechanism, which forms the fundamental building block of our network, FocusNet++. We employ a hybrid loss based on balanced cross entropy, Tversky loss and the adaptive logarithmic loss to enhance the performance along with fast convergence. Our results show that FocusNet++ achieves state-of-the-art results across various benchmark metrics for the ISIC 2018 melanoma segmentation and the cell nuclei segmentation datasets with fewer parameters and FLOPs.

KW - Group Ateention, Medical Image Segmentation, Residual Learning

U2 - 10.1109/ISBI48211.2021.9433918

DO - 10.1109/ISBI48211.2021.9433918

M3 - Conference contribution

BT - 2021 IEEE 18th International Symposium on Biomedical Imaging (ISBI)

Y2 - 13 April 2021 through 16 April 2021

ER -