Leveraging memory for improved medical image segmentation with limited parameters

Raffaele Berzoini; Marco D. Santambrogio; Eleonora D'Arnese

Leveraging memory for improved medical image segmentation with limited parameters

Raffaele Berzoini, Marco D. Santambrogio, Eleonora D'Arnese

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

Abstract

Deep Learning (DL) has emerged as a valuable solution for a fast and automatic segmentation limiting the burden on clinical personnel especially for segmentation. Although most imaging techniques produce 3D volumes, the most employed DL solutions work in a 2D fashion to keep hardware requirements limited and thus being applicable in real scenarios. Therefore, to exploit 3D information while keeping hardware requirements to the minimum we propose a novel architecture, the 2D Long Short Term Memory U-Net (2D LSTM U-Net). It combines the robust segmentation capabilities of a 2D U-Net with a volumetric understanding provided by the sequential data processing strengths of LSTM cells. Experimental results on the CT-ORG and BraTS 2020 datasets demonstrate the model’s effectiveness in binary and multi-class segmentation, achieving results on par with state-of-the-art exploiting 2D and 3D models despite having 1.6× and 16.5× less parameters, respectively.

Original language	English
Title of host publication	Proceedings of the 47th Annual International Conference of the IEEE Engineering in Medicine and Biology Society
Publisher	Institute of Electrical and Electronics Engineers
Pages	1-4
Number of pages	4
Publication status	Accepted/In press - 8 Apr 2025
Event	The 47th Annual International Conference of the IEEE Engineering in Medicine and Biology Society - Bella Center, Copenhagen, Denmark Duration: 14 Jul 2025 → 17 Jul 2025 Conference number: 47 https://embc.embs.org/2025/

Publication series

Name	Proceedings of the Annual International Conference of the IEEE Engineering in Medicine and Biology Society
Publisher	Institute of Electrical and Electronics Engineers
ISSN (Print)	1094-687X
ISSN (Electronic)	1558-4615

Conference

Conference	The 47th Annual International Conference of the IEEE Engineering in Medicine and Biology Society
Abbreviated title	EMBC 2025
Country/Territory	Denmark
City	Copenhagen
Period	14/07/25 → 17/07/25
Internet address	https://embc.embs.org/2025/

Keywords / Materials (for Non-textual outputs)

LSTM
segmentation
parameter limitation

Embargoed Document

BerzoiniEtalEMBC2025LeveragingMemoryforImproved
Accepted author manuscript, 1.41 MB
Licence: Creative Commons: Attribution (CC-BY)
Embargo ends: 17/07/25

Cite this

Berzoini, R., Santambrogio, M. D., & D'Arnese, E. (Accepted/In press). Leveraging memory for improved medical image segmentation with limited parameters. In Proceedings of the 47th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (pp. 1-4). (Proceedings of the Annual International Conference of the IEEE Engineering in Medicine and Biology Society). Institute of Electrical and Electronics Engineers.

Berzoini, Raffaele ; Santambrogio, Marco D. ; D'Arnese, Eleonora. / Leveraging memory for improved medical image segmentation with limited parameters. Proceedings of the 47th Annual International Conference of the IEEE Engineering in Medicine and Biology Society. Institute of Electrical and Electronics Engineers, 2025. pp. 1-4 (Proceedings of the Annual International Conference of the IEEE Engineering in Medicine and Biology Society).

@inproceedings{8fcd717b14a64b548863ae69e026b0ef,

title = "Leveraging memory for improved medical image segmentation with limited parameters",

abstract = "Deep Learning (DL) has emerged as a valuable solution for a fast and automatic segmentation limiting the burden on clinical personnel especially for segmentation. Although most imaging techniques produce 3D volumes, the most employed DL solutions work in a 2D fashion to keep hardware requirements limited and thus being applicable in real scenarios. Therefore, to exploit 3D information while keeping hardware requirements to the minimum we propose a novel architecture, the 2D Long Short Term Memory U-Net (2D LSTM U-Net). It combines the robust segmentation capabilities of a 2D U-Net with a volumetric understanding provided by the sequential data processing strengths of LSTM cells. Experimental results on the CT-ORG and BraTS 2020 datasets demonstrate the model{\textquoteright}s effectiveness in binary and multi-class segmentation, achieving results on par with state-of-the-art exploiting 2D and 3D models despite having 1.6× and 16.5× less parameters, respectively.",

keywords = "LSTM, segmentation, parameter limitation",

author = "Raffaele Berzoini and Santambrogio, {Marco D.} and Eleonora D'Arnese",

year = "2025",

month = apr,

day = "8",

language = "English",

series = "Proceedings of the Annual International Conference of the IEEE Engineering in Medicine and Biology Society",

publisher = "Institute of Electrical and Electronics Engineers",

pages = "1--4",

booktitle = "Proceedings of the 47th Annual International Conference of the IEEE Engineering in Medicine and Biology Society",

address = "United States",

note = "The 47th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, EMBC 2025 ; Conference date: 14-07-2025 Through 17-07-2025",

url = "https://embc.embs.org/2025/",

}

Berzoini, R, Santambrogio, MD & D'Arnese, E 2025, Leveraging memory for improved medical image segmentation with limited parameters. in Proceedings of the 47th Annual International Conference of the IEEE Engineering in Medicine and Biology Society. Proceedings of the Annual International Conference of the IEEE Engineering in Medicine and Biology Society, Institute of Electrical and Electronics Engineers, pp. 1-4, The 47th Annual International Conference of the IEEE Engineering in Medicine and Biology Society, Copenhagen, Denmark, 14/07/25.

Leveraging memory for improved medical image segmentation with limited parameters. / Berzoini, Raffaele; Santambrogio, Marco D.; D'Arnese, Eleonora.
Proceedings of the 47th Annual International Conference of the IEEE Engineering in Medicine and Biology Society. Institute of Electrical and Electronics Engineers, 2025. p. 1-4 (Proceedings of the Annual International Conference of the IEEE Engineering in Medicine and Biology Society).

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

TY - GEN

T1 - Leveraging memory for improved medical image segmentation with limited parameters

AU - Berzoini, Raffaele

AU - Santambrogio, Marco D.

AU - D'Arnese, Eleonora

N1 - Conference code: 47

PY - 2025/4/8

Y1 - 2025/4/8

N2 - Deep Learning (DL) has emerged as a valuable solution for a fast and automatic segmentation limiting the burden on clinical personnel especially for segmentation. Although most imaging techniques produce 3D volumes, the most employed DL solutions work in a 2D fashion to keep hardware requirements limited and thus being applicable in real scenarios. Therefore, to exploit 3D information while keeping hardware requirements to the minimum we propose a novel architecture, the 2D Long Short Term Memory U-Net (2D LSTM U-Net). It combines the robust segmentation capabilities of a 2D U-Net with a volumetric understanding provided by the sequential data processing strengths of LSTM cells. Experimental results on the CT-ORG and BraTS 2020 datasets demonstrate the model’s effectiveness in binary and multi-class segmentation, achieving results on par with state-of-the-art exploiting 2D and 3D models despite having 1.6× and 16.5× less parameters, respectively.

AB - Deep Learning (DL) has emerged as a valuable solution for a fast and automatic segmentation limiting the burden on clinical personnel especially for segmentation. Although most imaging techniques produce 3D volumes, the most employed DL solutions work in a 2D fashion to keep hardware requirements limited and thus being applicable in real scenarios. Therefore, to exploit 3D information while keeping hardware requirements to the minimum we propose a novel architecture, the 2D Long Short Term Memory U-Net (2D LSTM U-Net). It combines the robust segmentation capabilities of a 2D U-Net with a volumetric understanding provided by the sequential data processing strengths of LSTM cells. Experimental results on the CT-ORG and BraTS 2020 datasets demonstrate the model’s effectiveness in binary and multi-class segmentation, achieving results on par with state-of-the-art exploiting 2D and 3D models despite having 1.6× and 16.5× less parameters, respectively.

KW - LSTM

KW - segmentation

KW - parameter limitation

M3 - Conference contribution

T3 - Proceedings of the Annual International Conference of the IEEE Engineering in Medicine and Biology Society

SP - 1

EP - 4

BT - Proceedings of the 47th Annual International Conference of the IEEE Engineering in Medicine and Biology Society

PB - Institute of Electrical and Electronics Engineers

T2 - The 47th Annual International Conference of the IEEE Engineering in Medicine and Biology Society

Y2 - 14 July 2025 through 17 July 2025

ER -

Berzoini R, Santambrogio MD, D'Arnese E. Leveraging memory for improved medical image segmentation with limited parameters. In Proceedings of the 47th Annual International Conference of the IEEE Engineering in Medicine and Biology Society. Institute of Electrical and Electronics Engineers. 2025. p. 1-4. (Proceedings of the Annual International Conference of the IEEE Engineering in Medicine and Biology Society).

Leveraging memory for improved medical image segmentation with limited parameters

Abstract

Publication series

Conference

Keywords / Materials (for Non-textual outputs)

Embargoed Document

Fingerprint

Cite this