Can we explain machine learning-based prediction for rupture status assessments of intracranial aneurysms?

Although applying machine learning (ML) algorithms to rupture status assessment of intracranial aneurysms (IA) has yielded promising results, the opaqueness of some ML methods has limited their clinical translation. We presented the first explainability comparison of six commonly used ML algorithms: multivariate logistic regression (LR), support vector machine (SVM), random forest (RF), extreme gradient boosting (XGBoost), multi-layer perceptron neural network (MLPNN), and Bayesian additive regression trees (BART). A total of 112 IAs with known rupture status were selected for this study. The ML-based classification used two anatomical features, nine hemodynamic parameters, and thirteen morphologic variables. We utilized permutation feature importance, local interpretable model-agnostic explanations (LIME), and SHapley Additive exPlanations (SHAP) algorithms to explain and analyze 6 Ml algorithms. All models performed comparably: LR area under the curve (AUC) was 0.71; SVM AUC was 0.76; RF AUC was 0.73; XGBoost AUC was 0.78; MLPNN AUC was 0.73; BART AUC was 0.73. Our interpretability analysis demonstrated consistent results across all the methods; i.e., the utility of the top 12 features was broadly consistent. Furthermore, contributions of 9 important features (aneurysm area, aneurysm location, aneurysm type, wall shear stress maximum during systole, ostium area, the size ratio between aneurysm width, (parent) vessel diameter, one standard deviation among time-averaged low shear area, and one standard deviation of temporally averaged low shear area less than 0.4 Pa) were nearly the same. This research suggested that ML classifiers can provide explainable predictions consistent with general domain knowledge concerning IA rupture. With the improved understanding of ML algorithms, clinicians' trust in ML algorithms will be enhanced, accelerating their clinical translation.

4D-DSA: Development and Current Neurovascular Applications.

Originally described by Davis et al in 2013, 4D-Digital Subtraction Angiography (4D-DSA) has developed into a commercially available application of DSA in the angiography suite. 4D-DSA provides the user with 3D time-resolved images, allowing observation of a contrast bolus at any desired viewing angle through the vasculature and at any time point during the acquisition (any view at any time). 4D-DSA mitigates some limitations that are intrinsic to both 2D- and 3D-DSA images. The clinical applications for 4D-DSA include evaluations of AVMs and AVFs, intracranial aneurysms, and atherosclerotic occlusive disease. Recent advances in blood flow quantification using 4D-DSA indicate that these data provide both the velocity and geometric information necessary for the quantification of blood flow. In this review, we will discuss the development, acquisition, reconstruction, and current neurovascular applications of 4D-DSA volumes.

Optimizing the Quality of 4D-DSA Temporal Information.

BACKGROUND AND PURPOSE: Quantification of blood flow using a 4D-DSA would be useful in the diagnosis and treatment of cerebrovascular diseases. A protocol optimizing identification of density variations in the time-density curves of a 4D-DSA has not been defined. Our purpose was to determine the contrast injection protocol most likely to result in the optimal pulsatility signal strength. MATERIALS AND METHODS: Two 3D-printed patient-specific models were used and connected to a pulsatile pump and flow system, which delivered 250-260 mL/min to the model. Contrast medium (Isovue, 370 mg I/mL, 75% dilution) was injected through a 6F catheter positioned upstream from the inlet of the model. 4D-DSA acquisitions were performed for the following injection rates: 1.5, 2.0, 2.5, 3.0 and 3.5 mL/s for 8 seconds. To determine pulsatility, we analyzed the time-density curve at the inlets using the oscillation amplitude and a previously described numeric metric, the sideband ratio. Vascular geometry from 4D-DSA reconstructions was compared with ground truth and micro-CT measurements of the model. Dimensionless numbers that characterize hemodynamics, Reynolds and Craya-Curtet, were calculated for each injection rate. RESULTS: The strongest pulsatility signal occurred with the 2.5 mL/s injections. The largest oscillation amplitudes were found with 2.0- and 2.5-mL/s injections. Geometric accuracy was best preserved with injection rates of >1.5 mL/s. CONCLUSIONS: An injection rate of 2.5 mL/s provided the strongest pulsatility signal in the 4D-DSA time-density curve. Geometric accuracy was best preserved with injection rates above 1.5 mL/s. These results may be useful in future in vivo studies of blood flow quantification.

Quantification of Blood Velocity with 4D Digital Subtraction Angiography Using the Shifted Least-Squares Method.

BACKGROUND AND PURPOSE: 4D-DSA provides time-resolved 3D-DSA volumes with high temporal and spatial resolutions. The purpose of this study is to investigate a shifted least squares method to estimate the blood velocity from the 4D DSA images. Quantitative validation was performed using a flow phantom with an ultrasonic flow probe as ground truth. Quantification of blood velocity in human internal carotid arteries was compared with measurements generated from 3D phase-contrast MR imaging. MATERIALS AND METHODS: The centerlines of selected vascular segments and the time concentration curves of each voxel along the centerlines were determined from the 4D-DSA dataset. The temporal shift required to achieve a minimum difference between any point and other points along the centerline of a segment was calculated. The temporal shift as a function of centerline point position was fit to a straight line to generate the velocity. The proposed shifted least-squares method was first validated using a flow phantom study. Blood velocities were also estimated in the 14 ICAs of human subjects who had both 4D-DSA and phase-contrast MR imaging studies. Linear regression and correlation analysis were performed on both the phantom study and clinical study, respectively. RESULTS: Mean velocities of the flow phantom calculated from 4D-DSA matched very well with ultrasonic flow probe measurements with 11% relative root mean square error. Mean blood velocities of ICAs calculated from 4D-DSA correlated well with phase-contrast MR imaging measurements with Pearson correlation coefficient r = 0.835. CONCLUSIONS: The availability of 4D-DSA provides the opportunity to use the shifted least-squares method to estimate velocity in vessels within a 3D volume.

Natural history of the canine vein pouch aneurysm model.

BACKGROUND AND PURPOSE: The canine vein pouch aneurysm model is widely used for testing and development of devices directed at the endovascular treatment of aneurysms. Our purpose was to determine the incidence of spontaneous thrombosis and rupture of these aneurysms. MATERIALS AND METHODS: A retrospective review of laboratory records of canine vein pouch aneurysms made during a 6-year period was performed. The aneurysm and parent artery dimensions as well as incidences of spontaneous thrombosis and rupture were noted. RESULTS: During the interval studied, 326 vein patch aneurysms were made in 310 canines. Of these, 102 were sidewall (lateral) and 224 were bifurcation aneurysms. Spontaneous occlusion occurred in 9 of the sidewall aneurysms and in only 1 of the bifurcation aneurysms. None of the aneurysms ruptured. CONCLUSION: Spontaneous occlusion of the sidewall canine vein patch aneurysm occurred less than 10% of the time; in the bifurcation aneurysms, it almost never occurred. These characteristics enhance the value of this model for use in testing of devices intended for the endovascular treatment of aneurysms.

CT angiographic appearance of in-stent restenosis of intracranial arteries treated with the Wingspan stent.

Four patients underwent angioplasty and stenting of medically refractory symptomatic intracranial atherosclerosis with the new Wingspan stent system. In all 4 patients, CT angiography (CTA) showed an abnormality within the stented segment that was suggestive of nonocclusive in-stent thrombus. However, subsequent conventional angiography findings were typical for in-stent restenosis. The CTA imaging features of in-stent restenosis are important to recognize, and the misinterpretation of in-stent restenosis as in-stent thrombus may result in inappropriate management.

Dynamic Angiography and Perfusion Imaging Using Flat Detector CT in the Angiography Suite: A Pilot Study in Patients with Acute Middle Cerebral Artery Occlusions.

BACKGROUND AND PURPOSE: Perfusion and angiographic imaging using intravenous contrast application to evaluate stroke patients is now technically feasible by flat detector CT performed by the angiographic system. The aim of this pilot study was to show the feasibility and qualitative comparability of a novel flat detector CT dynamic perfusion and angiographic imaging protocol in comparison with a multimodal stroke MR imaging protocol. MATERIALS AND METHODS: In 12 patients with acute stroke, MR imaging and the novel flat detector CT protocol were performed before endovascular treatment. Perfusion parameter maps (MTT, TTP, CBV, CBF) and MIP/volume-rendering technique images obtained by using both modalities (MR imaging and flat detector CT) were compared. RESULTS: Comparison of MIP/volume-rendering technique images demonstrated equivalent visibility of the occlusion site. Qualitative comparison of perfusion parameter maps by using ASPECTS revealed high Pearson correlation coefficients for parameters CBF, MTT, and TTP (0.95-0.98), while for CBV, the coefficient was lower (0.49). CONCLUSIONS: We have shown the feasibility of a novel dynamic flat detector CT perfusion and angiographic protocol for the diagnosis and triage of patients with acute ischemic stroke. In a qualitative comparison, the parameter maps and MIP/volume-rendering technique images compared well with MR imaging. In our opinion, this flat detector CT application may substitute for multisection CT imaging in selected patients with acute stroke so that in the future, patients with acute stroke may be directly referred to the angiography suite, thereby avoiding transportation and saving time.

Evaluation of the TriSpan neck bridge device for the treatment of wide-necked aneurysms: an experimental study in canines.

BACKGROUND AND PURPOSE: Many wide-necked aneurysms are difficult or impossible to treat with the Guglielmi detachable coil (GDC). The purpose of this study was to evaluate the use of a neck bridging device, the TriSpan coil, in combination with standard GDCs for the treatment of wide-necked aneurysms in an experimental canine aneurysm model. METHODS: Of 24 experimental aneurysms in 12 animals, 19 (7 lateral and 12 terminal) were treated with the TriSpan coil in conjunction with standard GDCs. Digital subtraction angiography (DSA) was performed on all animals immediately after treatment. In 6 animals, follow-up DSA and histological evaluation were performed 4 weeks after treatment. In the remaining 6, DSA was done at both 90 and 180 days after treatment. Histological evaluation was done immediately after the 180-day angiographic evaluation. RESULTS: The TriSpan was easy to use in conjunction with the standard GDC. Because of their geometry, some lateral aneurysms were difficult or impossible to treat with this device. Greater than 90% aneurysm occlusion was obtained in all 19 aneurysms. In no instance was there evidence of coil migration, herniation, or aneurysm recanalization. Histological evaluation of the tissue surrounding the TriSpan coil showed tissue responses similar to that seen with standard GDCs. CONCLUSIONS: These results show that the TriSpan coil in conjunction with standard GDCs can be used safely and effectively for the treatment of wide-necked aneurysms in this canine model. Positioning and deployment of the neck bridge in aneurysms having an acute angle with the long axis of their parent artery are difficult or impossible. It is likely that this device, used in conjunction with the standard GDC, will allow treatment of some wide-necked aneurysms that are not treatable with the GDC alone.

Corticosteroid-induced mediastinal widening in myasthenia gravis.

Corticosteroid-induced mediastinal widening with simulated focal lymphadenopathy was found in a patient having myasthenia gravis. To my knowledge, this is the first report of such changes occurring in a patient having myasthenia gravis. As the use of corticosteroids in the treatment of myasthenia gravis becomes more common, this association can be expected to occur more often. Reemphasis of this phenomenon should help to avoid unnecessary surgical intervention in these patients.

Anatomic considerations for computed tomography of the optic chiasm.

When the clinical distinction between lesions of the optic nerves or the chiasm is not apparent, computed tomographic scans should be performed at both 0 degrees and 25 degrees or 35 degrees to the orbital-meatal line (Reid's baseline). The former plane is preferable for the demonstration of the optic nerves or orbital structures, while the latter allows superior visualization of perichiasmatic structures.

Digital subtraction arteriography. A new method for evaluation of extracranial occlusive disease.

Digital subtraction arteriography (DSA) allows visualization of both the intracranial and extracranial vasculature following an intravenous injection of contrast medium. One hundred consecutive patients were evaluated using this modality. In 32, DSA was compared with conventional arteriography. The degree of internal carotid artery stenosis and/or occlusion shown by DSA was confirmed by conventional arteriography in 27 cases (84%). Abnormalities of the extracranial vasculature was demonstrated by DSA in 60% of patients with suspected cerebrovascular disease. Digital subtraction arteriography demonstrated significant stenosis or occlusion of the carotid arteries in 30% of patients in whom Doppler and radionuclide studies were normal. Digital subtraction arteriography is useful in evaluation of the extracranial vasculature in a variety of clinical conditions. In some instances it serves as a substitute for conventional arteriography. However, its exact place in evaluation of extracranial vasculature disease remains to be defined.

Delayed visual loss due to trauma of the internal carotid artery.

The group of six patients in this study experienced delayed visual loss following head trauma. Visual loss occurred from 1 day to 13 years after the initial injury. All patients suffered indirect trauma to the internal carotid artery resulting in formation of either an aneurysm or pseudoaneurysm or a carotid-cavernous fistula. Review of the radiologic and clinical findings was performed in six patients. The diagnosis was established by computed tomography, magnetic resonance imaging, and angiography. All patients had follow-up clinical evaluation and imaging studies. Treatment by neurosurgical or interventional neuroradiologic procedures resulted in significant visual improvement in five patients. Different pathophysiologic mechanisms could be correlated with the delayed visual loss produced by the two types of lesions. The pathologic changes associated with the aneurysms/pseudoaneurysms included direct compression of optic nerves and/or chiasm and intracranial hematoma. A carotid-cavernous fistula caused delayed visual loss by either hematoma at the orbital apex or compression of the chiasm and/or optic nerves by saccular dilatation of the cavernous sinus. The delayed onset of decreased vision following head trauma should alert the physician to the possibility of a traumatic aneurysm/pseudoaneurysm or a carotid-cavernous fistula. Different neuro-ophthalmologic symptoms can usually be correlated with the pathologic changes demonstrated by neuroimaging procedures.

Spontaneous regression of optic gliomas: thirteen cases documented by serial neuroimaging.

OBJECTIVE: To demonstrate spontaneous regression of large, clinically symptomatic optic pathway gliomas in patients with and without neurofibromatosis type 1 (NF-1). METHODS: Patient cases were collected through surveys at 2 consecutive annual meetings of the North American Neuro-Ophthalmology Society (NANOS) and through requests on the NANOSNET Internet listserv. Serial documentation of tumor signal and size, using magnetic resonance imaging in 11 patients and computed tomography in 2 patients, was used to evaluate clinically symptomatic optic pathway gliomas. All tumors met radiologic criteria for the diagnosis of glioma and 4 patients had biopsy confirmation of their tumors. In 3 patients, some attempt at therapy had been made many years before regression occurred. In one of these, radiation treatment had been given 19 years before tumor regression, while in another, chemotherapy had been administered 5 years before signal changes in the tumor. In the third patient, minimal surgical debulking was performed 1 year before the tumor began to shrink. RESULTS: Spontaneous tumor shrinkage was noted in 12 patients. Eight patients did not have NF-1. In an additional patient without NF-1, a signal change within the tumor without associated shrinkage was detected. Tumor regression was associated with improvement in visual function in 10 of 13 patients, stability of function in 1, and deterioration in 2. CONCLUSIONS: Large, clinically symptomatic optic gliomas may undergo spontaneous regression. Regression was seen in patients with and without NF-1. Regression may manifest either as an overall shrinkage in tumor size, or as a signal change on magnetic resonance imaging. A variable degree of improvement in visual function may accompany regression. The possibility of spontaneous regression of an optic glioma should be considered in the planning of treatment of patients with these tumors.

Digital subtraction angiography and continuous-wave Doppler studies. Their use in postoperative study of patients with carotid endarterectomy.

Postoperative study of 40 patients who underwent carotid endarterectomy was performed using digital subtraction angiography (DSA) and transcutaneous continuous-wave Doppler (CWD) studies. The two techniques were comparable in defining recurrent stenosis in the reconstructed carotid arteries and progressive disease in the contralateral carotid artery. Digital subtraction angiography appeared to be more sensitive than the CWD method in detecting minor progression of carotid disease. Our data suggested that CWD and DSA have comparable ability to identify significant (greater than 50%) carotid artery stenosis in this group of patients. However, DSA provided more detailed imaging of the carotid system and may give enough information about progression of carotid disease to avoid preoperative arteriography in some patients.

A comparison of standard cerebral arteriography with noninvasive Doppler imaging and intravenous angiography.

Standard cervical carotid arteriography was performed on 36 patients and compared with results of noninvasive Doppler arterial imaging and intravenous angiography (IVA). The 72 carotid arteries were anatomically classified by standard angiograms as follows; group 1 (normal), 13 arteries; group 2 (wall disease [50% stenosis]), 14 arteries; group 3 (moderate stenosis [5-% to 75% stenosis]) 12 arteries; group 4 (severe stenosis [greater than 75%]), 15 arteries; and group 5 (complete occlusion), 18 arteries.

Computerized arteriography of the cerebrovascular system: its use with intravenous administration of contrast material.

A unique method of computerized image enhancement makes it possible to visualize the arterial system by intravenous (IV) injection of small doses of standard contrast agent. This technique has been used to study the intracranial and extracranial circulation of more than 100 patients. Occlusion, stenosis, aneurysmal change, plaquing, and ulceration can be identified by computerized IV arteriography. Failures (less than 10%) results from inadequate venous access, extravasation of dye, or patient motion. This technique avoids the need for arteriography. The major risk of IV arteriography is contrast medium reaction. There is a close correlation between IV and standard arteriographic images. Intravenous arteriography does not require hospitalization, is suited for repetitive testing, and has promise as a method of diagnostic screening for stroke prevention.

The vestibular aqueduct--tomographic evaluation in Meniere's disease--a preliminary report.

Frontal and off-lateral complex motion tomography was performed in 72 consecutive patients selected for endolymphatic sac surgery for episodic vertigo or sensorineural hearing loss. They were selected for this operation because of a fluctuating hearing loss. The morphology and visualization of the vestibular aqueduct and the degree of periaqueductal pneumatization were correlated with surgical results and the presence of bilateral disease. The surgical results were evaluated with use of the classification of the American Academy of Ophthalmology and Otolaryngology. Computer analysis was performed on the data for vestibular aqueduct radiographic anatomy and surgical results or bilateral disease. There was no correlation between visualization or morphology of the vestibular aqueduct and surgical results or presence of bilateral disease.

Aneurysm hemodynamics: an experimental study.

PURPOSE: To study the flow of blood in aneurysms. METHODS: A canine model was used to study the hemodynamics of lateral, bifurcation, and terminal aneurysms with angiography and color Doppler techniques. FINDINGS: Flow within experimental aneurysms, although not laminar, is seldom if ever turbulent, but rather is highly predictable, varying primarily according to the relationship of the aneurysm to its parent artery. CONCLUSIONS: These studies support earlier in vitro work and provide further evidence that not all aneurysms share similar stresses. A more complete understanding of these hemodynamic features will be useful in the establishment of criteria that allow recognition of those aneurysms that are more or less likely to rupture, to grow, or to thrombose.

Treatment of experimental canine carotid aneurysms with platinum coils.

PURPOSE: To evaluate and compare the deliverability, positioning, stability, and effectiveness of aneurysm occlusion and the incidence of parent-artery thrombosis of two different types of platinum coils, using a canine carotid aneurysm model. METHODS: 29 experimental canine carotid aneurysms (19 lateral, 6 bifurcation, and 4 terminal) were constructed and treated with complex-shaped fibered platinum coils and simple curved nonfibered platinum Guglielmi detachable coils (GDCs). RESULTS: Fibered complex coils were stable, producing 38% complete aneurysm occlusion and 61% average reduction in aneurysm lumen size but resulting in 19% parent artery occlusions. GDC coils were stable, producing 31% complete aneurysm occlusion and 95% average reduction in aneurysm lumen size with no parent-artery occlusions. CONCLUSIONS: GDC coils produced an average reduction in aneurysm lumen size of 95% without any associated parent-artery occlusions. There were no delayed migrations of GDC coils. The ability to remove, reposition, and detach a coil was the most significant feature of the GDC coil.

Canine model of terminal arterial aneurysm.

A canine model of terminal aneurysms using a venous pouch surgical technique is described. This model mimics the anatomy and hemodynamics of some types of carotid and basilar tip aneurysms. This technique produces aneurysms 15 x 21 mm in size. The aneurysms have been used to investigate the hemodynamics and treatment of terminal aneurysms.