Cardiac Fibrosis Automated Diagnosis Based on FibrosisNet Network Using CMR Ischemic Cardiomyopathy

• Published in: Diagnostics (Basel) 2024-01, Vol.14 (3), p.255
• Main Authors: Bekheet, Mohamed, Sallah, Mohammed, Alghamdi, Norah S, Rusu-Both, Roxana, Elgarayhi, Ahmed, Elmogy, Mohammed
• Format: Article
• Language: English
• Subjects: Accuracy; Algorithms; Artificial intelligence; Automation; Cardiac arrhythmia; Cardiac stress tests; Cardiomyopathy; Cardiovascular disease; Classification; CMR; Deep learning; Diagnostic imaging; Fibrosis; FibrosisNet; Health aspects; Heart; Heart failure; Histopathology; Ischemia; LGE; Machine learning; Magnetic resonance imaging; Medical imaging; Medical imaging equipment; Medical research; Methods; Myocardial ischemia; Neural networks; Scars; Tomography
• ISSN: 2075-4418; 2075-4418
• DOI: 10.3390/diagnostics14030255

Ischemic heart condition is one of the most prevalent causes of death that can be treated more effectively and lead to fewer fatalities if identified early. Heart muscle fibrosis affects the diastolic and systolic function of the heart and is linked to unfavorable cardiovascular outcomes. Cardiac magnetic resonance (CMR) scarring, a risk factor for ischemic heart disease, may be accurately identified by magnetic resonance imaging (MRI) to recognize fibrosis. In the past few decades, numerous methods based on MRI have been employed to identify and categorize cardiac fibrosis. Because they increase the therapeutic advantages and the likelihood that patients will survive, developing these approaches is essential and has significant medical benefits. A brand-new method that uses MRI has been suggested to help with diagnosing. Advances in deep learning (DL) networks contribute to the early and accurate diagnosis of heart muscle fibrosis. This study introduces a new deep network known as FibrosisNet, which detects and classifies fibrosis if it is present. It includes some of 17 various series layers to achieve the fibrosis detection target. The introduced classification system is trained and evaluated for the best performance results. In addition, deep transfer-learning models are applied to the different famous convolution neural networks to find fibrosis detection architectures. The FibrosisNet architecture achieves an accuracy of 96.05%, a sensitivity of 97.56%, and an F1-Score of 96.54%. The experimental results show that FibrosisNet has numerous benefits and produces higher results than current state-of-the-art methods and other advanced CNN approaches.