{
  "abstract": "Introduction Cardiomyocyte ageing and injury are essential contributors to cardiac dysfunction and heart failure, yet early detection remains a challenge due to the limitations of traditional assessment tech- niques. Existing methods, such as mechanosensory tracking and calcium imaging, are expensive, labour-intensive, and not readily adaptable for efficiently analyzing large volumes of cardiomyocyte data. Additionally, conventional motion analysis approaches often fail to capture localized contrac- tion variations, leading to an inaccurate classification of cardiomyocyte health. To address these challenges, we propose an advanced AI-driven framework that leverages Transformer-based mo- tion analysis for automated cardiomyocyte classification into Healthy, Aged, or Damaged states. It provides a comprehensive strategy and a high-accuracy classification system. The proposed method improves cardiomyocyte phenotyping by effectively differentiating contraction intensity patterns, enabling early identification of functional decline with superior accuracy and resilience over traditional motion tracking techniques. 1–3 Methodology The novel dataset was uniquely generated in a controlled biosafety environment at the Univer- sity of East London. Induced Pluripotent Stem cell-derived cardiomyocytes were thawed, counted (700,000–1M cells), and seeded in 96-well plates. Over 8 days, cells differentiated into beating heart cells, with their contractile motion recorded daily using high-resolution Vitro Imaging data and microscopy at 30–87 FPS. Motion tracking was performed from Day 1 (first visible contraction) to Day 8 (cessation of contractions), allowing temporal phenotyping of cardiomyocyte function.Motion tracking4–6 was performed using Optical Flow Farneback tracking to extract systolic (M systolic) and diastolic (M diastolic) motion intensities. Cells were classified as:Healthy: Msystolic ≥ 0.7, Mdiastolic ≥ 0.5Aged: 0.3 ≤ Msystolic < 0.7, 0.3 ≤ Mdiastolic < 0.5Damaged: Msystolic < 0.3, Mdiastolic < 0.3A Transformer-based Spatio-Temporal AI Model (TimeSformer) was trained alongside CNN-LSTM networks, improving classification accuracy. Manual bounding box annotations were applied us- ing Roboflow open source automation tools, and SHAP, LIME, GradCAM, Integrated gradient, DeepLift and attention maps enhanced model interpretability for clinical right decision. Kruskal- Wallis and Dunn’s post-hoc tests verified statistical significance.Results and Discussion On day 7, 8 cells are maximum damaged. However, novel find- ings suggest a biological and experimental cardiomyocyte maturation and degradation timeline. The logical explanation behind this variation is rooted in the stages of differentiation, functional maturation, and cellular stress over time.In terms of biological process in table 1 : Abstract BS45 Table 1Biological processes underlying cardiomyocyte aging and functional decline Day Range Logic behind the dyesfunction decline event Day Aged and healthy 1– 3, few • iPSC-derived cardiomyocytes do not immediately exhibit full contraction-relaxation cycles after differentiation.• Cells are still developing sarcomeric structures (the contractile unit of cardiomyocytes).• Calcium signaling pathways are not fully optimized, leading to irregular contractions. Day Healthy 4– 6, • By Day 5, cardiomyocytes have fully developed sarcomeres, allowing efficient contraction-relaxation cycles.• Calcium-handling proteins (RyR, SERCA2a) become fully functional, improving contractility.• Cells display synchronous beating, a hallmark of mature car- diomyocytes. Day Damaged 7– 8, • Cellular Aging: Prolonged culture without sufficient growth factors leads to cellular senescence.• Oxidative Stress Accumulation: Reactive oxygen species (ROS) damage cell membranes, causing fragility and loss of function.• Calcium Dysregulation: Impaired calcium reuptake leads to spontaneous arrhythmic contractions, fragmentation, and necrosis.• Cell Detachment and Apoptosis: Some cardiomyocytes un- dergo programmed cell death, resulting in debris and noise in microscopy images.The TimeSformer model achieved 94.2% accuracy , outperforming CNN-LSTM (89.6%) and traditional motion tracking (78.3% ). Self-attention mechanisms improved differentiation betweenAged and Damaged cardiomyocytes. Segmentation performance, evaluated via Dice Score (0.91) and IoU (0.87), demonstrated high agreement with expert labels. LIME and gradient attention mechanism in terms of novel transformer analysis confirmed that localized contraction intensity predicts cell health more than global motion magnitude.Conclusion This study establishes Transformer-based motion phenotyping as a precise tool for early cardiomyocyte dysfunction detection, enabling scalable and interpretable AI-driven as- sessment for cardiac disease research.1 Blockdiagram of Cardiomyocyte Aging and Injury Using Motion PhenotypingAdvanced Transformer-based AI framework for early detection and prediction of Cardiomyocyte aging and injury using motion phenotyping in figure 1 . Abstract BS45 Figure 1Proposed AI framework for cardiomyocyte phenotypingReferences Naiela E, et al. Early diagnosis of cardiovascular diseases in the era of artificial intelligence. An in-depth Review, Springer 2024;13:427. DOI: 0.7759/cureus.55869.Yidi Zhang, et al. A visual detection method of cardiomyocyte relaxation and contraction. AIP Advances 2023;13:025028. DOI: 10.1063/5.0133456.Sofia A, et al. (2023). Cardiac System during the Ageing Process. Ageing and Disease 2023;14:1105–1122. DOI: 10.14336/AD.2023.0115.Feridooni HA, Dibb KM, Howlett SE. How cardiomyocyte excitation, calcium release and contraction become altered with age. JMCC 2015;83:62–72. DOI: 10.1016/j.yjmcc.2014.12.004.Blair CA, Pruitt BL. Mechanobiology assays with applications in cardiomyocyte biology and cardiotoxicity. Advanced Healthcare Materials 2020; First published: 09 April. DOI: 10.1002/adhm.201901656.Ribeiro ASF, Zerolo BE, López-Espuela F, et al. Cardiac system during the ageing process. Aging Disease 2023;14:1105–22. DOI: 10.14336/AD.2023.0115.",
  "authors": [
    {
      "affiliations": [
        "Department of Engineering and Computing, School of Architecture, Computing and Engineering, University of East London, London, UK"
      ],
      "name": "Md Abu Sufian"
    },
    {
      "affiliations": [
        "Department of Engineering and Computing, School of Architecture, Computing and Engineering, University of East London, London, UK"
      ],
      "name": "Nadeem Qazi"
    },
    {
      "affiliations": [
        "Department of Bioscience, School of Health, Sport and Bioscience, University of East London, London, UK"
      ],
      "name": "Prashant Ruchaya"
    },
    {
      "affiliations": [
        "Department of Engineering and Computing, School of Architecture, Computing and Engineering, University of East London, London, UK"
      ],
      "name": "Nabeela Berardinelli"
    },
    {
      "affiliations": [
        "Department of Computer Science and Digital Technologies, School of Architecture, Computing and Engineering, University of East London, London, UK"
      ],
      "name": "Mustansar Ali Ghazanfar"
    }
  ],
  "title": "BS45 Advanced transformer-based AI framework for early detection and prediction of cardiomyocyte aging and injury using motion phenotyping",
  "uid": "2d2849c0-0b45-5ece-a4ed-97921378bb7f"
}