Detection and analysis of cerebral aneurysms based on X-ray rotational angiography - the CADA 2020 challenge

The STOIC2021 COVID-19 AI challenge: Applying reusable training methodologies to private data

Challenges drive the state-of-the-art of automated medical image analysis. The quantity of public training data that they provide can limit the performance of their solutions. Public access to the training methodology for these solutions remains absent. This study implements the Type Three (T3) challenge format, which allows for training solutions on private data and guarantees reusable training methodologies. With T3, challenge organizers train a codebase provided by the participants on sequestered training data. T3 was implemented in the STOIC2021 challenge, with the goal of predicting from a computed tomography (CT) scan whether subjects had a severe COVID-19 infection, defined as intubation or death within one month. STOIC2021 consisted of a Qualification phase, where participants developed challenge solutions using 2000 publicly available CT scans, and a Final phase, where participants submitted their training methodologies with which solutions were trained on CT scans of 9724 subjects. The organizers successfully trained six of the eight Final phase submissions. The submitted codebases for training and running inference were released publicly. The winning solution obtained an area under the receiver operating characteristic curve for discerning between severe and non-severe COVID-19 of 0.815. The Final phase solutions of all finalists improved upon their Qualification phase solutions.

Multimodal exploration of the intracranial aneurysm wall

PURPOSE

Intracranial aneurysms (IAs) are pathological changes of the intracranial vessel wall, although clinical image data can only show the vessel lumen. Histology can provide wall information but is typically restricted to ex vivo 2D slices where the shape of the tissue is altered.

METHODS

We developed a visual exploration pipeline for a comprehensive view of an IA. We extract multimodal information (like stain classification and segmentation of histologic images) and combine them via 2D to 3D mapping and virtual inflation of deformed tissue. Histological data, including four stains, micro-CT data and segmented calcifications as well as hemodynamic information like wall shear stress (WSS), are combined with the 3D model of the resected aneurysm.

RESULTS

Calcifications were mostly present in the tissue part with increased WSS. In the 3D model, an area of increased wall thickness was identified and correlated to histology, where the Oil red O (ORO) stained images showed a lipid accumulation and the alpha-smooth muscle actin (aSMA) stained images showed a slight loss of muscle cells.

CONCLUSION

Our visual exploration pipeline combines multimodal information about the aneurysm wall to improve the understanding of wall changes and IA development. The user can identify regions and correlate how hemodynamic forces, e.g. WSS, are reflected by histological structures of the vessel wall, wall thickness and calcifications.

StrokeNet: An automated approach for segmentation and rupture risk prediction of intracranial aneurysm

Intracranial Aneurysms (IA) present a complex challenge for neurosurgeons as the risks associated with surgical intervention, such as Subarachnoid Hemorrhage (SAH) mortality and morbidity, may outweigh the benefits of aneurysmal occlusion in some cases. Hence, there is a critical need for developing techniques that assist physicians in assessing the risk of aneurysm rupture to determine which aneurysms require treatment. However, a reliable IA rupture risk prediction technique is currently unavailable. To address this issue, this study proposes a novel approach for aneurysm segmentation and multidisciplinary rupture prediction using 2D Digital Subtraction Angiography (DSA) images. The proposed method involves training a fully connected convolutional neural network (CNN) to segment aneurysm regions in DSA images, followed by extracting and fusing different features using a multidisciplinary approach, including deep features, geometrical features, Fourier descriptor, and shear pressure on the aneurysm wall. The proposed method also adopts a fast correlation-based filter approach to drop highly correlated features from the set of fused features. Finally, the selected fused features are passed through a Decision Tree classifier to predict the rupture severity of the associated aneurysm into four classes: Mild, Moderate, Severe, and Critical. The proposed method is evaluated on a newly developed DSA image dataset and on public datasets to assess its generalizability. The system's performance is also evaluated on DSA images annotated by expert neurosurgeons for the rupture risk assessment of the segmented aneurysm. The proposed system outperforms existing state-of-the-art segmentation methods, achieving an 85 % accuracy against annotated DSA images for the risk assessment of aneurysmal rupture.

Deep-learning based detection of vessel occlusions on CT-angiography in patients with suspected acute ischemic stroke

Swift diagnosis and treatment play a decisive role in the clinical outcome of patients with acute ischemic stroke (AIS), and computer-aided diagnosis (CAD) systems can accelerate the underlying diagnostic processes. Here, we developed an artificial neural network (ANN) which allows automated detection of abnormal vessel findings without any a-priori restrictions and in <2 minutes. Pseudo-prospective external validation was performed in consecutive patients with suspected AIS from 4 different hospitals during a 6-month timeframe and demonstrated high sensitivity (≥87%) and negative predictive value (≥93%). Benchmarking against two CE- and FDA-approved software solutions showed significantly higher performance for our ANN with improvements of 25-45% for sensitivity and 4-11% for NPV (p ≤ 0.003 each). We provide an imaging platform ( https://stroke.neuroAI-HD.org ) for online processing of medical imaging data with the developed ANN, including provisions for data crowdsourcing, which will allow continuous refinements and serve as a blueprint to build robust and generalizable AI algorithms.

USING CONVOLUTIONAL NEURAL NETWORK-BASED SEGMENTATION FOR IMAGE-BASED COMPUTATIONAL FLUID DYNAMICS SIMULATIONS OF BRAIN ANEURYSMS: INITIAL EXPERIENCE IN AUTOMATED MODEL CREATION

"Image-based" computational fluid dynamics (CFD) simulations provide insights into each patient's hemodynamic environment. However, current standard procedures for creating CFD models start with manual segmentation and are time-consuming, hindering the clinical translation of image-based CFD simulations. This feasibility study adopts deep-learning-based image segmentation (hereafter referred to as Artificial Intelligence (AI) segmentation) to replace manual segmentation to accelerate CFD model creation. Two published convolutional neural network-based AI methods (MIScnn and DeepMedic) were selected to perform CFD model extraction from three-dimensional (3D) rotational angiography data containing intracranial aneurysms. In this study, aneurysm morphological and hemodynamic results using models generated by AI segmentation methods were compared with those obtained by two human users for the same data. Interclass coefficients (ICC), Bland-Altman plots, and Pearson's correlation coefficients (PCC) were combined to assess how well AI-generated CFD models were performed. We found that almost perfect agreement was obtained between the human and AI results for all eleven morphological and five out of eight hemodynamic parameters, while a moderate agreement was obtained from the remaining three hemodynamic parameters. Given this level of agreement, using AI segmentation to create CFD models is feasible, given more developments.

Modelling blood flow in patients with heart valve disease using deep learning: A computationally efficient method to expand diagnostic capabilities in clinical routine

Introduction

The computational modelling of blood flow is known to provide vital hemodynamic parameters for diagnosis and treatment-support for patients with valvular heart disease. However, most diagnosis/treatment-support solutions based on flow modelling proposed utilize time- and resource-intensive computational fluid dynamics (CFD) and are therefore difficult to implement into clinical practice. In contrast, deep learning (DL) algorithms provide results quickly with little need for computational power. Thus, modelling blood flow with DL instead of CFD may substantially enhances the usability of flow modelling-based diagnosis/treatment support in clinical routine. In this study, we propose a DL-based approach to compute pressure and wall-shear-stress (WSS) in the aorta and aortic valve of patients with aortic stenosis (AS).

Methods

A total of 103 individual surface models of the aorta and aortic valve were constructed from computed tomography data of AS patients. Based on these surface models, a total of 267 patient-specific, steady-state CFD simulations of aortic flow under various flow rates were performed. Using this simulation data, an artificial neural network (ANN) was trained to compute spatially resolved pressure and WSS using a centerline-based representation. An unseen test subset of 23 cases was used to compare both methods.

Results

ANN and CFD-based computations agreed well with a median relative difference between both methods of 6.0% for pressure and 4.9% for wall-shear-stress. Demonstrating the ability of DL to compute clinically relevant hemodynamic parameters for AS patients, this work presents a possible solution to facilitate the introduction of modelling-based treatment support into clinical practice.

Aneurysm Neck Overestimation has a Relatively Modest Impact on Simulated Hemodynamics

INTRODUCTION

Overestimation of intracranial aneurysm neck width by 3D angiography is a recognized clinical problem, and has long been a concern for image-based computational fluid dynamics (CFD). Recently, it was demonstrated that neck overestimation in 3D rotational angiography (3DRA) could be corrected via segmentation with upsampled resolution and gradient enhancement (SURGE). Our aim was to leverage this approach to determine whether and how neck overestimation actually impacts CFD-derived hemodynamics.

MATERIALS AND METHODS

A subset of 17 cases having the largest neck errors from a consecutive clinical sample of 60 was segmented from 3DRA using both standard watershed and SURGE methods. High-fidelity, pulsatile CFD was performed, and a variety of scalar hemodynamic parameters that have been associated with aneurysm growth and/or rupture status were derived.

RESULTS

With a few exceptions, flow and wall shear stress (WSS) patterns were qualitatively similar between neck-overestimated and corrected models. Sac-averaged WSS values were significantly lower after neck correction (p = 0.0005) but were highly correlated with their neck-overestimated counterparts (R² = 0.98). Jet impingement was significantly more concentrated in the neck-corrected vs. -uncorrected models (p = 0.0011), and only moderately correlated (R² = 0.61). Parameters quantifying velocity or WSS fluctuations were not significantly different after neck correction, but this reflected their poorer correlations (R² < 0.4). Nevertheless, for all hemodynamic parameters, median absolute differences were < 26%, and no parameter had more than 5/17 cases with absolute differences > 50%.

CONCLUSION

Differences in hemodynamics due to neck width overestimation were found to be at most equal to, and often less than, those reported for other sources of error/uncertainty in intracranial aneurysm CFD, such as solver settings or assumed inflow rates.

Geometric uncertainty in intracranial aneurysm rupture status discrimination: a two-site retrospective study

OBJECTIVES

Assessing the risk associated with unruptured intracranial aneurysms (IAs) is essential in clinical decision making. Several geometric risk parameters have been proposed for this purpose. However, performance of these parameters has been inconsistent. This study evaluates the performance and robustness of geometric risk parameters on two datasets and compare it to the uncertainty inherent in assessing these parameters and quantifies interparameter correlations.

METHODS

Two datasets containing 244 ruptured and unruptured IA geometries from 178 patients were retrospectively analysed. IAs were stratified by anatomical region, based on the PHASES score locations. 37 geometric risk parameters representing four groups (size, neck, non-dimensional, and curvature parameters) were assessed. Analysis included standardised absolute group differences (SADs) between ruptured and unruptured IAs, ratios of SAD to median relative uncertainty (MRU) associated with the parameters, and interparameter correlation.

RESULTS

The ratio of SAD to MRU was lower for higher dimensional size parameters (ie, areas and volumes) than for one-dimensional size parameters. Non-dimensional size parameters performed comparatively well with regard to SAD and MRU. SAD was higher in the posterior anatomical region. Correlation of parameters was strongest within parameter (sub)groups and between size and curvature parameters, while anatomical region did not strongly affect correlation patterns.

CONCLUSION

Non-dimensional parameters and few parameters from other groups were comparatively robust, suggesting that they might generalise better to other datasets. The data on discriminative performance and interparameter correlations presented in this study may aid in developing and choosing robust geometric parameters for use in rupture risk models.

A deep learning-based automatic system for intracranial aneurysms diagnosis on three-dimensional digital subtraction angiographic images

BACKGROUND

Intracranial aneurysms (IAs) are a life-threatening disease. Their rupture can lead to hemorrhagic stroke. Most studies applying deep learning for the detection of aneurysms are based on angiographic images. However, critical diagnostic information such as morphology and aneurysm location are not captured by deep learning algorithms and still require manual assessments.

PURPOSE

Digital subtraction angiography (DSA) is the gold standard for aneurysm diagnosis. To facilitate the fully automatic diagnosis of aneurysms, we proposed a comprehensive system for the detection, morphology measurement, and location classification of aneurysms on three-dimensional DSA images, allowing automatic diagnosis without further human input.

METHODS

The system comprised three neural networks: a network for aneurysm detection, a network for morphology measurement, and a network for aneurysm location identification. A cross-scale dual-path transformer module was proposed to effectively fuse local and global information to capture aneurysms of varying sizes. A multitask learning approach was also proposed to allow an accurate localization of aneurysm neck for morphology measurement.

RESULTS

The cross-scale dual-path transformer module was shown to outperform other state-of-the-art network architectures, improving segmentation, and classification accuracy. The detection network in our system achieved an F2 score of 0.946 (recall 93%, precision 100%), better than the winning team in the Cerebral Aneurysm Detection and Analysis challenge. The measurement network achieved a relative error of less than 10% for morphology measurement, at the same level as human operators. Perfect accuracy (100%) was achieved on aneurysm location classification.

CONCLUSIONS

We have demonstrated that a comprehensive system can automatically detect, measure morphology and report the aneurysm location of aneurysms without human intervention. This can be a potential tool for the diagnosis of IAs, improving radiologists' performance and reducing their workload.

Improving visualization of three-dimensional aneurysm features via segmentation with upsampled resolution and gradient enhancement (SURGE)

BACKGROUND

Intracranial aneurysm neck width tends to be overestimated when measured with three-dimensional rotational angiography (3DRA) compared with two-dimensional digital subtraction angiography (2D-DSA), owing to high curvature at the neck. This may affect morphological and hemodynamic analysis in support of treatment planning. We present and validate a method for extracting high curvature features, such as aneurysm ostia, during segmentation of 3DRA images.

METHODS

In our novel SURGE (segmentation with upsampled resolution and gradient enhancement) approach, the gradient of an upsampled image is sharpened before gradient-based watershed segmentation. Neck measurements were performed for both standard and SURGE segmentations of 3DRA for 60 consecutive patients and compared with those from 2D-DSA. Those segmentations were also qualitatively compared for surface topology and morphology.

RESULTS

Compared with the standard watershed method, SURGE reduced neck measurement error relative to 2D-DSA by >60%: median error was 0.49 mm versus 0.17 mm for SURGE, which is less than the average pixel resolution (~0.33 mm) of the 3DRA dataset. SURGE reduced neck width overestimations >1 mm from 13/60 to 5/60 cases. Relative to 2D-DSA, standard segmentations were overestimated by 16% and 93% at median and 95th percentiles, respectively, compared with only 6% and 37%, respectively, for SURGE.

CONCLUSION

SURGE provides operators with high-level control of the image gradient, allowing recovery of high-curvature features such as aneurysm ostia from 3DRA where conventional algorithms may fail. Compared with standard segmentation and tedious manual editing, SURGE provides a faster, easier, and more objective method for assessing aneurysm ostia and morphology.

Morphology-aware multi-source fusion-based intracranial aneurysms rupture prediction

OBJECTIVES

We proposed a new approach to train deep learning model for aneurysm rupture prediction which only uses a limited amount of labeled data.

METHOD

Using segmented aneurysm mask as input, a backbone model was pretrained using a self-supervised method to learn deep embeddings of aneurysm morphology from 947 unlabeled cases of angiographic images. Subsequently, the backbone model was finetuned using 120 labeled cases with known rupture status. Clinical information was integrated with deep embeddings to further improve prediction performance. The proposed model was compared with radiomics and conventional morphology models in prediction performance. An assistive diagnosis system was also developed based on the model and was tested with five neurosurgeons.

RESULT

Our method achieved an area under the receiver operating characteristic curve (AUC) of 0.823, outperforming deep learning model trained from scratch (0.787). By integrating with clinical information, the proposed model's performance was further improved to AUC = 0.853, making the results significantly better than model based on radiomics (AUC = 0.805, p = 0.007) or model based on conventional morphology parameters (AUC = 0.766, p = 0.001). Our model also achieved the highest sensitivity, PPV, NPV, and accuracy among the others. Neurosurgeons' prediction performance was improved from AUC=0.877 to 0.945 (p = 0.037) with the assistive diagnosis system.

CONCLUSION

Our proposed method could develop competitive deep learning model for rupture prediction using only a limited amount of data. The assistive diagnosis system could be useful for neurosurgeons to predict rupture.

KEY POINTS

• A self-supervised learning method was proposed to mitigate the data-hungry issue of deep learning, enabling training deep neural network with a limited amount of data. • Using the proposed method, deep embeddings were extracted to represent intracranial aneurysm morphology. Prediction model based on deep embeddings was significantly better than conventional morphology model and radiomics model. • An assistive diagnosis system was developed using deep embeddings for case-based reasoning, which was shown to significantly improve neurosurgeons' performance to predict rupture.