Using a Convolutional Neural Network and Convolutional Long Short-term Memory to Automatically Detect Aneurysms on 2D-DSA Images (Preprint)

BACKGROUND It is hard to distinguish cerebral aneurysms from overlapping vessels based on 2D DSA images due to their lack of spatial information. OBJECTIVE The aim of this study was to construct a deep learning diagnostic system to improve the ability to detect posterior communicating artery (PCoA) aneurysms on 2D-DSA images and validate the efficiency of the deep learning diagnostic system in 2D-DSA aneurysm detection. METHODS We proposed a two-stage detecting system. First, we established the regional localization stage (RLS) to automatically locate specific detection regions of raw 2D-DSA sequences. Then, in the intracranial aneurysm detection stage (IADS), we constructed the Bi-input+RetinaNet+C-LSTM framework to compare the performance of aneurysm detection with the existing three frameworks. Each of the frameworks had a fivefold cross-validation scheme. The area under the curve (AUC), the receiver operating characteristic (ROC) curve, and mean average precision (mAP) were used to validate the efficiency of different frameworks. The sensitivity, specificity and accuracy were used to identify the abilities of different frameworks. RESULTS A total of 255 patients with PCoA aneurysms and 20 patients without aneurysms were included in this study. The best AUC results of RetinaNet, RetinaNet+C-LSTM, Bi-input+RetinaNet and Bi-input+RetinaNet+C-LSTM were 0.95, 0.96, 0.92 and 0.97, respectively. The sensitivities of RetinaNet, RetinaNet+C-LSTM, Bi-input+RetinaNet, Bi-input+RetinaNet+C-LSTM, and human experts were 89.00% (67.02% to 98.43%), 88.00% (65.76% to 98.06%), 87.00% (64.53% to 97.66%), 89.00% (67.02% to 98.43%), and 90% (68.30% to 98.77%), respectively. The specificity of RetinaNet, RetinaNet+C-LSTM, Bi-input+RetinaNet, Bi-input+RetinaNet+C-LSTM, and human expert were 80.00% (56.34% to 94.27%), 89.00% (67.02% to 98.43%), 86.00% (63.31% to 97.24%), 93.00% (72.30% to 99.56%), and 90% (68.30% to 98.77%), respectively. The accuracies of RetinaNet, RetinaNet+C-LSTM, Bi-input+RetinaNet, Bi-input+RetinaNet+C-LSTM, and human experts were 84.50% (69.57% to 93.97%), 88.50% (74.44% to 96.39%), 86.50% (71.97% to 95.22%), 91.00% (77.63% to 97.72%), and 90.00% (76.34% to 97.21%), respectively. CONCLUSIONS A two-stage aneurysm detection system can reduce the time cost and the computational load. According to our results, more spatial and temporal information can help improve the performances of the frameworks so that Bi-input+RetinaNet+C-LSTM has the best performance compared to the other frameworks. Our study demonstrates that our system can assist doctors in detecting intracranial aneurysms on 2D-DSA images.

Value of 3D-DSA in the detection of intracranial aneurysms

Aim: To determine the value of three-dimensional (3D) digital subtraction angiography (DSA) in the detection of intracranial aneurysms and to compare 3D technique with DSA. Materials and Methods: A retrospective analysis of 50 patients with 60 intracranial aneurysms who underwent both conventional DSA and 3D-DSA for the evaluation of intracranial aneurysms was conducted. The presence of aneurysms, detection of aneurysmal neck, size, location, presence of additional and small aneurysms analyzed from the two protocols were compared. Results: Three-dimensional technique detected 54 aneurysms while conventional DSA detected 38 aneurysms. There was no correlation between aneurysm detection and aneurysm neck detection in the two technologies observed, but there was a difference in detection performance depending on the technology used. Three-dimensional technique detected 52 aneurysm necks while conventional DSA detected 24 aneurysm necks. There was a statistically significant and positive relationship between the detected size of the aneurysm using 3D technique and DSA technology. Three-dimensional technique detected 24 additional aneurysms while conventional DSA detected only six additional aneurysms. Conclusions: Three-dimensional technique are more successful in the detection of aneurysms, their necks and small aneurysms in comparison to digital subtraction angiography, but difference is not statistically significant. The size of the aneurysm statistically significant affects the aneurysm neck detection by conventional DSA.

Randomized study comparing 3D virtual reality and conventional 2D on-screen teaching of cerebrovascular anatomy

OBJECTIVE Performing aneurysmal clipping requires years of training to successfully understand the 3D neurovascular anatomy. This training has traditionally been obtained by learning through observation. Currently, with fewer operative aneurysm clippings, stricter work-hour regulations, and increased patient safety concerns, novel teaching methods are required for young neurosurgeons. Virtual-reality (VR) models offer the opportunity to either train a specific surgical skill or prepare for an individual surgery. With this study, the authors aimed to compare the spatial orientation between traditional 2D images and 3D VR models in neurosurgical residents or medical students. METHODS Residents and students were each randomly assigned to describe 4 aneurysm cases, which could be either 2D images or 3D VR models. The time to aneurysm detection as well as a spatial anatomical description was assessed via an online questionnaire and compared between the groups. The aneurysm cases were 10 selected patient cases treated at the authors’ institution. RESULTS Overall, the time to aneurysm detection was shorter in the 3D VR model compared to 2D images, with a trend toward statistical significance (25.77 ± 37.26 vs 45.70 ± 51.94 seconds, p = 0.052). No significant difference was observed for residents (3D VR 24.47 ± 40.16 vs 2D 33.52 ± 56.06 seconds, p = 0.564), while in students a significantly shorter time to aneurysm detection was measured using 3D VR models (26.95 ± 35.39 vs 59.16 ± 44.60 seconds, p = 0.015). No significant differences between the modalities for anatomical and descriptive spatial mistakes were observed. Most participants (90%) preferred the 3D VR models for aneurysm detection and description, and only 1 participant (5%) described VR-related side effects such as dizziness or nausea. CONCLUSIONS VR platforms facilitate aneurysm recognition and understanding of its spatial anatomy, which could make them the preferred method compared to 2D images in the years to come.

Using convolutional neural network and long short time memory to automatically detect aneurysm on 2D DSA images (Preprint)

BACKGROUND It is hard to distinguish cerebral aneurysm from overlap vessels based on the 2D DSA images, for its lack the spatial information. OBJECTIVE The aim of this study is to construct a deep learning diagnostic system to improve the ability of detecting the PCoA aneurysm on 2D-DSA images and validate the efficiency of deep learning diagnostic system in 2D-DSA aneurysm detecting. METHODS We proposed a two stage detecting system. First, we established the regional localization stage (RLS) to automatically locate specific detection region of raw 2D-DSA sequences. And then, in the intracranial aneurysm detection stage (IADS) ,we build three different frames, RetinaNet, RetinaNet+LSTM, Bi-input+RetinaNet+LSTM, to detect the aneurysms. Each of the frame had fivefold cross-validation scheme. The area under curve (AUC), the receiver operating characteristic (ROC) curve, and mean average precision (mAP) were used to validate the efficiency of different frames. The sensitivity, specificity and accuracy were used to identify the ability of different frames. RESULTS 255 patients with PCoA aneurysms and 20 patients without aneurysm were included in this study. The best results of AUC of the RetinaNet, RetinaNet+LSTM, and Bi-input+RetinaNet+LSTM were 0.95, 0.96, and 0.97, respectively. The sensitivity of the RetinaNet, RetinaNet+LSTM, and Bi-input+RetinaNet+LSTM were 81.65% (59.40% to 94.76%), 87.91% (64.24% to 98.27%), 84.50% (69.57% to 93.97%), respectively. The specificity of the RetinaNet, RetinaNet+LSTM, and Bi-input+RetinaNet+LSTM were 88.89% (66.73% to 98.41%), 88.12% (66.06% to 98.08%), and 88.50% (74.44% to 96.39%), respectively. The accuracy of the RetinaNet, RetinaNet+LSTM, and Bi-input+RetinaNet+LSTM were 92.71% (71.29% to 99.54%), 89.42% (68.13% to 98.49%), and 91.00% (77.63% to 97.72%), respectively. CONCLUSIONS Two stage aneurysm detecting system can reduce time cost and the computation load. According to our results, more spatial and temporal information can help improve the performance of the frames, so that Bi-input+RetinaNet+LSTM has the best performance compared to other frames. And our study can demonstrate that our system was feasible to assist doctor to detect intracranial aneurysm on 2D-DSA images.

Deep neural network-based detection and segmentation of intracranial aneurysms on 3D rotational DSA

Objective Accurate diagnosis and measurement of intracranial aneurysms are challenging. This study aimed to develop a 3D convolutional neural network (CNN) model to detect and segment intracranial aneurysms (IA) on 3D rotational DSA (3D-RA) images. Methods 3D-RA images were collected and annotated by 5 neuroradiologists. The annotated images were then divided into three datasets: training, validation, and test. A 3D Dense-UNet-like CNN (3D-Dense-UNet) segmentation algorithm was constructed and trained using the training dataset. Diagnostic performance to detect aneurysms and segmentation accuracy was assessed for the final model on the test dataset using the free-response receiver operating characteristic (FROC). Finally, the CNN-inferred maximum diameter was compared against expert measurements by Pearson’s correlation and Bland-Altman limits of agreement (LOA). Results A total of 451 patients with 3D-RA images were split into n = 347/41/63 training/validation/test datasets, respectively. For aneurysm detection, observed FROC analysis showed that the model managed to attain a sensitivity of 0.710 at 0.159 false positives (FP)/case, and 0.986 at 1.49 FP/case. The proposed method had good agreement with reference manual aneurysmal maximum diameter measurements (8.3 ± 4.3 mm vs. 7.8 ± 4.8 mm), with a correlation coefficient r = 0.77, small bias of 0.24 mm, and LOA of -6.2 to 5.71 mm. 37.0% and 77% of diameter measurements were within ±1 mm and ±2.5 mm of expert measurements. Conclusions A 3D-Dense-UNet model can detect and segment aneurysms with relatively high accuracy using 3D-RA images. The automatically measured maximum diameter has potential clinical application value.