A lightweight neural network with multiscale feature enhancement for liver CT segmentation

Mohammed Yusuf Ansari; Yin Yang; Shidin Balakrishnan; Julien Abinahed; Abdulla Al-Ansari; Mohamed Warfa; Omran Almokdad; Ali Barah; Ahmed Omer; Ajay Vikram Singh; Pramod Kumar Meher; Jolly Bhadra; Osama Halabi; Mohammad Farid Azampour; Nassir Navab; Thomas Wendler; Sarada Prasad Dakua

doi:10.1038/s41598-022-16828-6

A lightweight neural network with multiscale feature enhancement for liver CT segmentation

Mohammed Yusuf Ansari, Yin Yang, Shidin Balakrishnan, Julien Abinahed, Abdulla Al-Ansari, Mohamed Warfa, Omran Almokdad, Ali Barah, Ahmed Omer, Ajay Vikram Singh, Pramod Kumar Meher, Jolly Bhadra, Osama Halabi, Mohammad Farid Azampour, Nassir Navab, Thomas Wendler, Sarada Prasad Dakua

Informatics 16 - Chair of Computer Aided Medical Procedures

Research output: Contribution to journal › Article › peer-review

56 Scopus citations

Abstract

Segmentation of abdominal Computed Tomography (CT) scan is essential for analyzing, diagnosing, and treating visceral organ diseases (e.g., hepatocellular carcinoma). This paper proposes a novel neural network (Res-PAC-UNet) that employs a fixed-width residual UNet backbone and Pyramid Atrous Convolutions, providing a low disk utilization method for precise liver CT segmentation. The proposed network is trained on medical segmentation decathlon dataset using a modified surface loss function. Additionally, we evaluate its quantitative and qualitative performance; the Res16-PAC-UNet achieves a Dice coefficient of 0.950 ± 0.019 with less than half a million parameters. Alternatively, the Res32-PAC-UNet obtains a Dice coefficient of 0.958 ± 0.015 with an acceptable parameter count of approximately 1.2 million.

Original language	English
Article number	14153
Journal	Scientific Reports
Volume	12
Issue number	1
DOIs	https://doi.org/10.1038/s41598-022-16828-6
State	Published - Dec 2022

Access to Document

10.1038/s41598-022-16828-6

Cite this

Ansari, M. Y., Yang, Y., Balakrishnan, S., Abinahed, J., Al-Ansari, A., Warfa, M., Almokdad, O., Barah, A., Omer, A., Singh, A. V., Meher, P. K., Bhadra, J., Halabi, O., Azampour, M. F., Navab, N., Wendler, T., & Dakua, S. P. (2022). A lightweight neural network with multiscale feature enhancement for liver CT segmentation. Scientific Reports, 12(1), Article 14153. https://doi.org/10.1038/s41598-022-16828-6

@article{536e64c03b974e4bbfa9481e33a48201,

title = "A lightweight neural network with multiscale feature enhancement for liver CT segmentation",

abstract = "Segmentation of abdominal Computed Tomography (CT) scan is essential for analyzing, diagnosing, and treating visceral organ diseases (e.g., hepatocellular carcinoma). This paper proposes a novel neural network (Res-PAC-UNet) that employs a fixed-width residual UNet backbone and Pyramid Atrous Convolutions, providing a low disk utilization method for precise liver CT segmentation. The proposed network is trained on medical segmentation decathlon dataset using a modified surface loss function. Additionally, we evaluate its quantitative and qualitative performance; the Res16-PAC-UNet achieves a Dice coefficient of 0.950 ± 0.019 with less than half a million parameters. Alternatively, the Res32-PAC-UNet obtains a Dice coefficient of 0.958 ± 0.015 with an acceptable parameter count of approximately 1.2 million.",

author = "Ansari, {Mohammed Yusuf} and Yin Yang and Shidin Balakrishnan and Julien Abinahed and Abdulla Al-Ansari and Mohamed Warfa and Omran Almokdad and Ali Barah and Ahmed Omer and Singh, {Ajay Vikram} and Meher, {Pramod Kumar} and Jolly Bhadra and Osama Halabi and Azampour, {Mohammad Farid} and Nassir Navab and Thomas Wendler and Dakua, {Sarada Prasad}",

note = "Publisher Copyright: {\textcopyright} 2022, The Author(s).",

year = "2022",

month = dec,

doi = "10.1038/s41598-022-16828-6",

language = "English",

volume = "12",

journal = "Scientific Reports",

issn = "2045-2322",

publisher = "Nature Publishing Group",

number = "1",

}

Ansari, MY, Yang, Y, Balakrishnan, S, Abinahed, J, Al-Ansari, A, Warfa, M, Almokdad, O, Barah, A, Omer, A, Singh, AV, Meher, PK, Bhadra, J, Halabi, O, Azampour, MF, Navab, N, Wendler, T & Dakua, SP 2022, 'A lightweight neural network with multiscale feature enhancement for liver CT segmentation', Scientific Reports, vol. 12, no. 1, 14153. https://doi.org/10.1038/s41598-022-16828-6

TY - JOUR

T1 - A lightweight neural network with multiscale feature enhancement for liver CT segmentation

AU - Ansari, Mohammed Yusuf

AU - Yang, Yin

AU - Balakrishnan, Shidin

AU - Abinahed, Julien

AU - Al-Ansari, Abdulla

AU - Warfa, Mohamed

AU - Almokdad, Omran

AU - Barah, Ali

AU - Omer, Ahmed

AU - Singh, Ajay Vikram

AU - Meher, Pramod Kumar

AU - Bhadra, Jolly

AU - Halabi, Osama

AU - Azampour, Mohammad Farid

AU - Navab, Nassir

AU - Wendler, Thomas

AU - Dakua, Sarada Prasad

PY - 2022/12

Y1 - 2022/12

N2 - Segmentation of abdominal Computed Tomography (CT) scan is essential for analyzing, diagnosing, and treating visceral organ diseases (e.g., hepatocellular carcinoma). This paper proposes a novel neural network (Res-PAC-UNet) that employs a fixed-width residual UNet backbone and Pyramid Atrous Convolutions, providing a low disk utilization method for precise liver CT segmentation. The proposed network is trained on medical segmentation decathlon dataset using a modified surface loss function. Additionally, we evaluate its quantitative and qualitative performance; the Res16-PAC-UNet achieves a Dice coefficient of 0.950 ± 0.019 with less than half a million parameters. Alternatively, the Res32-PAC-UNet obtains a Dice coefficient of 0.958 ± 0.015 with an acceptable parameter count of approximately 1.2 million.

AB - Segmentation of abdominal Computed Tomography (CT) scan is essential for analyzing, diagnosing, and treating visceral organ diseases (e.g., hepatocellular carcinoma). This paper proposes a novel neural network (Res-PAC-UNet) that employs a fixed-width residual UNet backbone and Pyramid Atrous Convolutions, providing a low disk utilization method for precise liver CT segmentation. The proposed network is trained on medical segmentation decathlon dataset using a modified surface loss function. Additionally, we evaluate its quantitative and qualitative performance; the Res16-PAC-UNet achieves a Dice coefficient of 0.950 ± 0.019 with less than half a million parameters. Alternatively, the Res32-PAC-UNet obtains a Dice coefficient of 0.958 ± 0.015 with an acceptable parameter count of approximately 1.2 million.

UR - http://www.scopus.com/inward/record.url?scp=85136923717&partnerID=8YFLogxK

U2 - 10.1038/s41598-022-16828-6

DO - 10.1038/s41598-022-16828-6

M3 - Article

C2 - 35986015

AN - SCOPUS:85136923717

SN - 2045-2322

VL - 12

JO - Scientific Reports

JF - Scientific Reports

IS - 1

M1 - 14153

ER -

A lightweight neural network with multiscale feature enhancement for liver CT segmentation

Abstract

Access to Document

Other files and links

Fingerprint

Cite this