In the Netherlands, cardiovascular disease is the leading cause of death for citizens above 50 years old. To treat the disease, intravascular imaging is used to guide minimally-invasive interventional procedures resulting in improved outcomes, and also providing a better understanding of the disease progression. However, current studies focused only on pathology-inspired biomarkers which are limited by human interpretation, and only provide partial pieces of knowledge on the disease. In this project, we are looking for event-driven biomarkers of the disease to gain a full picture of the disease progression. The developed pipeline will train a deep convolutional neural network to end-to-end predict directly from images the probability of the occurrence of cardiac events in a certain time interval. To ensure that the highly relevant features are learned, we will develop a strategy using advanced interpretable ML tools to optimize the training of the model. We will refine the training to learn a robust set of imaging features by inspecting the learning efficiency of existing human-defined biomarkers and the significant instances. Using the quantifying result, we will modify the training strategies, and update data involvement. Once this step is achieved, the model is reliable to show the deep biomarkers as highly active regions using techniques such as the salience maps. Analyzing the biomarkers will contribute to a better understanding of the disease, such that patients can be treated specifically and personalized. The robust predicting model itself can also contribute to an optimal treatment and review strategies of patients.