Prediction of major adverse cardiovascular events using artificial intelligence and intracoronary optical coherence tomography

Coronary artery disease (CAD) is a pathological condition resulting from the progressive development of lipid-rich atherosclerotic plaques within the coronary artery walls, extending into the vessel lumen. This process involves chronic inflammation and gradual arterial narrowing, which can significantly impair blood flow and lead to severe clinical complications. Besides, plaque progressive growth may weaken the vessel wall, compromising its structural integrity. Given these threats, developing an effective risk prediction tool could facilitate early identification of high-risk cases, enabling preventing interventions and reducing the burden on the healthcare system. Intravascular optical coherence tomography (IVOCT) is an advanced imaging technique that has enhanced atherosclerotic lesion detection and assessment due to its exceptional resolution (10-20 μm). It uses a light-based system and requires the insertion of a catheter into the target vessel to analyse atherosclerotic plaques, highlighting their structural features and vulnerabilities. Despite its ability to provide detailed visualization of vessel morphology and lesions, interpreting IVOCT images remains challenging for healthcare professionals and requires specialized training. Artificial Intelligence (AI) offers a promising solution to enhance image analysis by automatically extracting crucial information, thereby supporting the complex task of cardiovascular risk prediction. By applying AI-based methodologies, this thesis work investigates the influence of OCT-derived features and clinical features on the occurrence of major adverse cardiovascular events (MACE). Specifically, the first step involves identifying a robust algorithm for effective classification across the available dataset. The second step applies this model to specific patient subsets to further explore potential correlations between clinical and OCT features and MACE occurrence in individuals with particular medical conditions. To achieve the above-mentioned objectives, a logistic regression algorithm was employed, proven suitable for this context. Its robustness was assessed through iterative feature selections and classifications across various patient combinations, evaluating average results and identifying the most informative features. Moreover, the general model’s performance on specific test set subgroups was estimated to examine its effectiveness in classifying patient subsets. Subsequently, subgroup analyses were performed on the whole dataset to evaluate the potential of developing an efficient model based exclusively on patient subcategories. The general model demonstrated fair overall performance, resulting appropriate for identifying patients with MACE across the entire dataset. Key metrics, including precision, recall, F1 score, and area under the receiver operating characteristic curve, were assessed. While some subset analyses were challenging due to dataset limitations, such as reduced patient population and imbalance between MACE and non-MACE cases, this approach provided deeper insights into the relationship between specific features (including IVOCT features) and cardiovascular complications. In conclusion, this work investigated the impact of clinical factors and atherosclerotic plaque characteristics extracted from IVOCT on MACE occurrence. Future improvements, such as integrating additional patient subsets or refining model precision, could contribute to the development of an effective AI-based cardiovascular risk prediction tool.