Neural Architecture Search Using Trajectory Metaheuristics to Classify Coronary Stenosis

Coronary stenosis is a disease that claims millions of lives each year. Early detection of this condition is crucial for patient survival. Currently, physicians perform detection by x-ray angiograms, however, the variability of diagnoses and the difficulty of access to expertise has led to the need for automated, computer-assisted diagnosis. In this work explores the use of deep learning to classify stenosis or non-stenosis in angiogram images using convolutional neural networks from scratch. A methodology to fine-tuning network architectures automatically using metaheuristic optimization techniques is proposed, demonstrating superior performance to fine-tuning empirically and proposing a new architecture in the literature to classify coronary stenosis. The proposed architectures achieved 86.02% and 95.67% F1-score with simulated annealing and iterated local search techniques, respectively.