Parameter-Free Bio-Inspired Channel Attention for Enhanced Cardiac MRI Reconstruction
Anam Hashmi, Julia Dietlmeier, Kathleen M. Curran, Noel E. O’Connor · 2025
Read Parameter-Free Bio-Inspired Channel Attention for Enhanced Cardiac MRI Reconstruction by Anam Hashmi, Julia Dietlmeier, Kathleen M. Curran, Noel E. O’Connor (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2505-23872.txt.
Introduction
Anam Hashmi, Julia Dietlmeier, Kathleen Curran, and Noel O'Connor borrow an idea from ecology to sharpen image software for heart scans. Rebuilding a clear heart image from limited scan data is a standing MRI challenge. Their tool is an attention module, part of a neural network that boosts useful features and damps the rest. Most attention modules rest on intuition rather than theory, and they wanted a firmer base.
They turned to a nonlinear equation describing how a single species grows in nature. From it they built an attention module needing no learned parameters. They report it beats the best parameter-free rivals on cardiac MRI reconstruction. The broader claim is that ecology, which has refined its growth laws over decades, can guide design of leaner and more principled network parts. The study applies the module to heart imaging as the test case.
Catalog record
| CREATOR | Anam Hashmi, Julia Dietlmeier, Kathleen M. Curran, Noel E. O’Connor |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 6 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2505-23872
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Parameter-Free Bio-Inspired Channel Attention for Enhanced Cardiac MRI Reconstruction
Anam Hashmi, Julia Dietlmeier, Kathleen M. Curran, Noel E. O’Connor
Abstract
Attention is a fundamental component of the human visual recognition system. The inclusion of attention in a convolutional neural network amplifies relevant visual features and suppresses the less important ones. Integrating attention mechanisms into convolutional neural networks enhances model performance and interpretability. Spatial and channel attention mechanisms have shown significant advantages across many downstream tasks in medical imaging. While existing attention modules have proven to be effective, their design often lacks a robust theoretical underpinning. In this study, we address this gap by proposing a non-linear attention architecture for cardiac MRI reconstruction and hypothesize that insights from ecological principles can guide the development of effective and efficient attention mechanisms. Specifically, we investigate a non-linear ecological difference equation that describes single-species population growth to devise a parameter-free attention module surpassing current state-of-the-art parameter-free methods.
1
Background and Motivation
Attention mechanisms play a pivotal role in enhancing the efficacy of convolutional neural networks (CNNs) by focusing on relevant visual features. Attention mechanisms have been successfully integrated as plug-and-play modules into CNNs for various applications such as medical image classification
[
9
]
, segmentation
[
8
]
, explainability
[
10
]
and most recently Cardiac Magnetic Resonance (CMR) reconstruction
[
5
,
2
]
. Several prominent architectures include the squeeze-and-excitation networks (SE)
[
11
]
, linear context transform block (LCT)
[
12
]
, and convolutional block attention module (CBAM)
[
13
]
. These and other state-of-the-art attention architectures
[
14
,
15
,
16
,
17
]
have demonstrated success in different applications, each tailored to specific tasks and requirements. While these attention architectures, along with others, exhibit promising results and some are even parameter-free
[
15
,
17
]
, their design often lacks a solid theoretical foundation. Thus, despite their efficacy, there remains a need for a more rigorous theoretical basis to underpin their design choices.
2
Biologically-Inspired Attention
In this work, we draw inspiration from non-linear ecological difference equations used in population biology
[
4
]
. These mathematical equations describe the dynamical system of growth of a given population depending on several environmental factors. Specifically, the following Equation (1)
[
4
]
(Fig.1 shows implementation in the blue box) describes single-species population growth:
N_{t+1}=\lambda[1+\alpha N_{t}]^{-b}N_{t}
(1)
N_{t+1}=\lambda[1+\alpha N_{t1}]^{-b}N_{t2}
(2)
Figure 1:
A schematic of the proposed biologically-inspired attention block. Our systematic design is based on modelling a non-linear ecological difference equation
[
4
]
in addition to average pooling, sigmoid activation function and L2-normalization layers. This follows the original Eq. (1) but with two separate branches
N_{t1}
and
N_{t2}
as defined in Eq. (2). Note that
\alpha=2
and
b=2
. Optimizing the hyperparameter values could yield better results, but is beyond the scope of this work.
The logic behind using non-linear ecological equations to model attention draws upon the principles of dynamic adaptation observed in natural systems. Biological populations dynamically adapt to environmental changes, emphasizing relevant factors and suppressing less important ones, similar to the objective of attention mechanisms in neural networks. Moreover, ecological systems, like population dynamics, often show complex non-linear behavior and integrating these dynamics allows to capture the real-world complexity more accurately. Thus, we hypothesize that non-linear ecological equations provide a framework for capturing the complex dynamic interactions between different visual features.
3
Methodology and Experiments
We experiment with the recently released CMRxRecon dataset
[
6
]
and obtain quantitative validation metrics such as
PSNR
,
MSE
and
SSIM
[
7
]
. For the reconstruction task, we utilize a CMR reconstruction network from
[
5
]
and term the configuration without attention layers as the
baseline
model. Our methodology follows
[
2
,
5
]
. The preprocessing and training details mirror those in
[
5
]
. Our computing pipeline was implemented using Python 3.11.5 and PyTorch 2.1.1.
Table 1:
Comparison of quantitative results. Wilcoxon signed-rank test was conducted to compare the SSIM scores of our method and the competing method SimAM. The results indicated a significant difference between the two methods (p-value < 0.001), demonstrating that our method significantly outperforms SimAM.
Attention
Computational overhead
PSNR
\uparrow
MSE
\downarrow
SSIM
\uparrow
baseline
0
36.2068
0.0002878
0.9245
+ SE
[
11
]
119,936
37.2863
0.0002355
0.9429
+ LCT
[
12
]
6,848
36.7562
0.0002618
0.9450
+ AB
[
14
]
3,424
34.3633
0.0004299
0.9357
+ ECA
[
16
]
90
37.9982
0.0002262
0.9527
+ SimAM
[
15
]
0
37.0492
0.0002583
0.9443
+ GCT
[
17
]
0
36.5874
0.0002695
0.9408
+ Proposed
0
37.7724
0.0002231
0.9496
Figure 2:
The reconstruction results and normalized absolute error maps for a long-axis cine image (case P1) with a ×10 acceleration factor, across different methods.
4
Conclusion
Our approach, based on non-linear ecological difference equations drawn from established ecological principles, outperformed existing parameter-free methods underscoring its effectiveness in enhancing cardiac MRI reconstruction.
Disclosure of Interests.
The authors have no competing interests to declare that are