Artificial neural networks: Predicting head CT findings in elderly patients presenting with minor head injury after a fall

作者：Dusenberry Michael W^*; Brown Charles K; Brewer Kori L

来源：American Journal of Emergency Medicine, 2017, 35(2): 260-267.

DOI：10.1016/j.ajem.2016.10.065

• Neural Network Models
• Elderly
• Head Injury
• Minor
• Falls

Objectives: To construct an artificial neural network (ANN) model that can predict the presence of acute CT findings with both high sensitivity and high specificity when applied to the population of patients >= age 65 years who have incurred minor head injury after a fall.
Methods: An ANN was created in the Python programming language using a population of 514 patients >= age 65 years presenting to the ED with minor head injury after a fall. The patient dataset was divided into three parts: 60% for "training", 20% for "cross validation", and 20% for "testing". Sensitivity, specificity, positive and negative predictive values, and accuracy were determined by comparing the model's predictions to the actual correct answers for each patient.
Results: On the "cross validation" data, the model attained a sensitivity ("recall") of 100.00%, specificity of 78.95%, PPV ("precision") of 78.95%, NPV of 100.00%, and accuracy of 88.24% in detecting the presence of positive head CTs. On the "test" data, the model attained a sensitivity of 97.78%, specificity of 89.47%, PPV of 88.00%, NPV of 98.08%, and accuracy of 93.14% in detecting the presence of positive head CTs.
Conclusions: ANNs show great potential for predicting CT findings in the population of patients >= 65 years of age presenting with minor head injury after a fall. As a good first step, the ANN showed comparable sensitivity, predictive values, and accuracy, with a much higher specificity than the existing decision rules in clinical usage for predicting head CTs with acute intracranial findings.

• 出版日期2017-2
• 单位East Carolina