Development Of A Deep Learning Computed Tomography Angiography Model For Diagnosis Of Intracranial Aneurysms

BackgroundIntracranial aneurysm is a common cerebrovascular disease,and a common cause of cerebrovascular accidents.Timely and accurate diagnosis is of great significance for the prevention of intracranial aneurysm brain accident and early treatment.Computed Tomography Angiography(CTA)is the primary method for first-line diagnosis of intracranial aneurysms.CTA requires manual judgment,inevitably with some misdiagnosis,misdiagnosis,and individual judgment differences,and is timeconsuming and sometimes requires further diagnosis using more sensitive and specific digital minus angiography(Digital Subtraction Angiography,DSA).Computer-based in-depth learning-based auxiliary algorithms have been used in the field of medical imaging diagnostics to help improve the sensitivity and specificity of disease diagnosis.In this study,a CTA diagnostic model based on computer deep learning was proposed to improve the clinical effectiveness of intracranial aneurysm diagnosis.MethodsThe subjects were patients with clinically suspected intracranial aneurysms and craniofacial CTA examination,and two tertiary A-level general hospitals in a provincial capital city were the research site to collect clinical data and image images of patients who were admitted to hospital and had cranial CTA examination in 2012-2019.Methods for establishing CTA diagnostic model were developed based on deep learning algorithm:First,10%of all cases are randomly selected,after excluding cases not tested for gold standard diagnosis(DSA or surgery),the remaining gold standard cases are used as samples for validating the model,and the remaining cases,together with the aforementioned cases not tested by DSA,are used to train CTA image segmentation diagnostic model based on U-net convolutional neural network algorithm.Evaluation of diagnostic effectiveness of deep learning CTA diagnostic model:Using the method of epidemiological diagnostic test evaluation,the evaluation index used include sensitivity,specificity,accuracy,Youden’ index,positive prediction value,negative prediction value,under curve area.ResultsA total of 2412 cases of suspected intracranial aneurysms were collected,of which 2410 cases were eligible for inclusion,and 1372210 frame images were obtained,an average of 569 frames per patient.Of 2410 patients,43 per cent were male,57 per cent were female.Median age was 60(40-91 years),and average age was 61.6 years(±12.2 years).There were 834 CTA-positive cases(34.6 per cent)and 1,576 CTA-negative cases(65.4 per cent).Compared with the gold standard test,the depth learning-assisted CTA diagnostic model established in this study had a sensitivity of 0.80(95%CI,0.73-0.86)and a specificity of 0.88(95%CI,0.79-79-1)to verify the diagnosis of intracranial aneurysms in cases 0.95)respectively.The Youden index was 0.68(95%CI 0.52-0.80)The accuracy was 0.83(95%CI,0.77-87.For male and subarachnoid hemorrhage patients,the positive prediction values diagnosed in this model reached 0.97(95%CI,0.88-1.00)and 0.99(95%CI,0.82-1.00)respectively,with an Area Under Receiver Operating Characteristic Curve of 0.93(95%CI,0.89-0.96);for intracranial aneurysms≤3mm,the sensitivity of diagnosis is 0.86(95%CI,0.76-0.93),and one positive case can be detected for each 1.4 person diagnosed.ConclusionsIn the diagnosis of intracranial aneurysms,the CTA diagnostic model based on deep learning had high specificity and positive prediction values compared with DSA diagnosis.Compared with DSA,CTA is non-invasive examination,time-saving and has few adverse reactions,so the diagnosis of CTA in this model has potential clinical application value,and can be used as an auxiliary diagnostic method for suspected intracranial aneurysm cases,thus reducing use of invasive diagnosis.