Alibakhshi, Alireza, Jevsikov, Jevgeni, Ufumaka, Isreal, Binmadi, Rana, NG, Tiffany, Stowell, Catherine C., Hussain, Arshian, Abdi, Abas, Serej, Nasim Dadashi, Shun-Shin, Matthew J., Francis, Darrel P., Manisty, Charlotte H. and Zolgharni, Massoud ORCID: https://orcid.org/0000-0003-0904-2904 (2026) Phase-Guided quality regression in Apical Echocardiography via CNN-LSTM: a systematic backbone benchmarking study. In: Artificial Intelligence in Healthcare. Lecture Notes in Computer Science, 16877 (6877). Springer, pp. 17-30.

PDF ZolgharniM_Phase-guided quality regression in Apical_AM.pdf - Accepted Version Restricted to Repository staff only until 12 August 2027. Available under License Creative Commons Attribution. Download (3MB)

Abstract

Apical view acquisition errors in echocardiography can lead to inaccurate estimation of clinically important measurements, including ventricular volume, ejection fraction, and strain. In this work, we propose a phase-centred CNN-LSTM framework for automated regression of apical image quality in echocardiographic studies, investigating which backbone architecture can most effectively predict a continuous reduction in left ventricular long-axis length without relying on explicit mask-based measurements. A patient-specific reference standard was generated by first identifying end-diastole using a phase detection model and then measuring LV long-axis length. For each patient, the scan with the maximum observed LV length served as the patient-specific reference, and deviations across other scans were quantified using a heuristic formulation based on relative length reduction. Using this systematic labelling strategy, eight CNN-LSTM backbone architectures were trained and compared on 1,799 echocardiographic studies from 100 patients, comprising apical four-chamber and apical two-chamber views, with the goal of directly regressing the degree of length reduction from raw image sequences alone. Across backbone comparisons, convolution-based models showed the strongest overall performance, with EfficientNetV2 achieving the lowest regression error for apical four-chamber views and ConvNeXt-Tiny and ResNet50-level models showing competitive performance in apical two-chamber views. To assess clinical relevance, two expert-annotated evaluation datasets, each labelled by 10 experts, were used for ranking-based comparison against expert consensus. The proposed AI approach demonstrated agreement with expert consensus at a level comparable to several human experts and showed meaningful ability to identify scans associated with overestimation and underestimation of key echocardiographic measurements. These findings support the potential of backbone-driven regression of apical view quality as a scalable, mask-free, and clinically interpretable tool for echocardiographic quality control.

Item Type:	Book Chapter or Section
Identifier:	10.1007/978-3-032-35393-1_2
Additional Information:	Book chapter of a conference paper presented in the International Conference on AI in Healthcare (AIiH), 28-28 August, London, United Kingdom.
Subjects:	Computing > Intelligent systems Medicine and health
Related URLs:	Organisation
Date Deposited:	24 Sep 2026
Dates:	Date Publication status 22 May 2026 Accepted 12 August 2026 Published Online
School, department or research centre:	School of Computing and Engineering THRIVE (The Centre for Translational Healthcare Research, Innovation, Vision, and Excellence)
URI:	https://repository.uwl.ac.uk/id/eprint/15317

Actions (admin access)