A Machine Learning Approach for Predicting the Size of Abdominal Aortic Aneurysm

Abdominal aortic aneurysm (AAA) is a potentially life-threatening condition in which the main blood vessel that supplies blood to the rest of the body becomes abnormally enlarged. The gold standard in assessing AAA risk is via measuring AAA size from ultrasound or computed tomography images. However, when there is no imaging equipment available, clinicians rely on physical examinations to detect AAA, which is reported to have sensitivity of 68% only. The objective of this study is to investigate if machine learning models can reliably detect AAA using AAA risk factors. A total of 369 patients with 20 clinical features were collected. Two machine learning models were developed to differentiate three types of AAA (Normal AAA: < 3 cm, Small AAA: 3–5.4 cm and Large AAA: > 5.4 cm). Stage 1 model is trained to detect the presence of AAA, differentiating between normal and abnormal AAA (AAA size < 3 cm vs ≥ 3 cm). Stage 2 model is trained to differentiate between small and large AAA. The models went through extensive training and experimentation and yielded balanced accuracies of 83% and 66%, sensitivities of 83% and 66% and specificities of 82% and 79% for Stages 1 and 2, respectively. Compared to physician examination, our Stage 1 model shows a 15% improvement in detecting AAA. This study demonstrated that it is feasible to detect AAA using easily available clinical features, which can be useful for clinicians who have limited access to imaging facilities and benefits patients from the rural areas.