Deep learning for cerebral angiography segmentation from non-contrast computed tomography

Deep learning for 3D vascular segmentation in hierarchical phase contrast tomography: a case study on kidney

Automated blood vessel segmentation is critical for biomedical image analysis, as vessel morphology changes are associated with numerous pathologies. Still, precise segmentation is difficult due to the complexity of vascular structures, anatomical variations across patients, the scarcity of annotated public datasets, and the quality of images. Our goal is to provide a foundation on the topic and identify a robust baseline model for application to vascular segmentation using a new imaging modality, Hierarchical Phase-Contrast Tomography (HiP-CT). We begin with an extensive review of current machine-learning approaches for vascular segmentation across various organs. Our work introduces a meticulously curated training dataset, verified by double annotators, consisting of vascular data from three kidneys imaged using HiP-CT as part of the Human Organ Atlas Project. HiP-CT pioneered at the European Synchrotron Radiation Facility in 2020, revolutionizes 3D organ imaging by offering a resolution of around 20 μm/voxel and enabling highly detailed localised zooms up to 1-2 μm/voxel without physical sectioning. We leverage the nnU-Net framework to evaluate model performance on this high-resolution dataset, using both known and novel samples, and implementing metrics tailored for vascular structures. Our comprehensive review and empirical analysis on HiP-CT data sets a new standard for evaluating machine learning models in high-resolution organ imaging. Our three experiments yielded Dice similarity coefficient (DSC) scores of 0.9523, 0.9410, and 0.8585, respectively. Nevertheless, DSC primarily assesses voxel-to-voxel concordance, overlooking several crucial characteristics of the vessels and should not be the sole metric for deciding the performance of vascular segmentation. Our results show that while segmentations yielded reasonably high scores-such as centerline DSC ranging from 0.82 to 0.88, certain errors persisted. Specifically, large vessels that collapsed due to the lack of hydrostatic pressure (HiP-CT is an ex vivo technique) were segmented poorly. Moreover, decreased connectivity in finer vessels and higher segmentation errors at vessel boundaries were observed. Such errors, particularly in significant vessels, obstruct the understanding of the structures by interrupting vascular tree connectivity. Our study establishes the benchmark across various evaluation metrics, for vascular segmentation of HiP-CT imaging data, an imaging technology that has the potential to substantively shift our understanding of human vascular networks.

Deep Learning for 3D Vascular Segmentation in Phase Contrast Tomography

Automated blood vessel segmentation is critical for biomedical image analysis, as vessel morphology changes are associated with numerous pathologies. Still, precise segmentation is difficult due to the complexity of vascular structures, anatomical variations across patients, the scarcity of annotated public datasets, and the quality of images. Our goal is to provide a foundation on the topic and identify a robust baseline model for application to vascular segmentation using a new imaging modality, Hierarchical Phase-Contrast Tomography (HiP-CT). We begin with an extensive review of current machine learning approaches for vascular segmentation across various organs. Our work introduces a meticulously curated training dataset, verified by double annotators, consisting of vascular data from three kidneys imaged using Hierarchical Phase-Contrast Tomography (HiP-CT) as part of the Human Organ Atlas Project. HiP-CT, pioneered at the European Synchrotron Radiation Facility in 2020, revolutionizes 3D organ imaging by offering resolution around 20μm/voxel, and enabling highly detailed localized zooms up to 1μm/voxel without physical sectioning. We leverage the nnU-Net framework to evaluate model performance on this high-resolution dataset, using both known and novel samples, and implementing metrics tailored for vascular structures. Our comprehensive review and empirical analysis on HiP-CT data sets a new standard for evaluating machine learning models in high-resolution organ imaging. Our three experiments yielded Dice scores of 0.9523 and 0.9410, and 0.8585, respectively. Nevertheless, DSC primarily assesses voxel-to-voxel concordance, overlooking several crucial characteristics of the vessels and should not be the sole metric for deciding the performance of vascular segmentation. Our results show that while segmentations yielded reasonably high scores-such as centerline Dice values ranging from 0.82 to 0.88, certain errors persisted. Specifically, large vessels that collapsed due to the lack of hydro-static pressure (HiP-CT is an ex vivo technique) were segmented poorly. Moreover, decreased connectivity in finer vessels and higher segmentation errors at vessel boundaries were observed. Such errors, particularly in significant vessels, obstruct the understanding of the structures by interrupting vascular tree connectivity. Through our review and outputs, we aim to set a benchmark for subsequent model evaluations using various modalities, especially with the HiP-CT imaging database.

Automated anatomical labeling of the intracranial arteries via deep learning in computed tomography angiography

Background and purpose: Anatomical labeling of the cerebral vasculature is a crucial topic in determining the morphological nature and characterizing the vital variations of vessels, yet precise labeling of the intracranial arteries is time-consuming and challenging, given anatomical structural variability and surging imaging data. We present a U-Net-based deep learning (DL) model to automatically label detailed anatomical segments in computed tomography angiography (CTA) for the first time. The trained DL algorithm was further tested on a clinically relevant set for the localization of intracranial aneurysms (IAs). Methods: 457 examinations with varying degrees of arterial stenosis were used to train, validate, and test the model, aiming to automatically label 42 segments of the intracranial arteries [e.g., 7 segments of the internal carotid artery (ICA)]. Evaluation metrics included Dice similarity coefficient (DSC), mean surface distance (MSD), and Hausdorff distance (HD). Additionally, 96 examinations containing at least one IA were enrolled to assess the model's potential in enhancing clinicians' precision in IA localization. A total of 5 clinicians with different experience levels participated as readers in the clinical experiment and identified the precise location of IA without and with algorithm assistance, where there was a washout period of 14 days between two interpretations. The diagnostic accuracy, time, and mean interrater agreement (Fleiss' Kappa) were calculated to assess the differences in clinical performance of clinicians. Results: The proposed model exhibited notable labeling performance on 42 segments that included 7 anatomical segments of ICA, with the mean DSC of 0.88, MSD of 0.82 mm and HD of 6.59 mm. Furthermore, the model demonstrated superior labeling performance in healthy subjects compared to patients with stenosis (DSC: 0.91 vs. 0.89, p < 0.05; HD: 4.75 vs. 6.19, p < 0.05). Concurrently, clinicians with model predictions achieved significant improvements when interpreting the precise location of IA. The clinicians' mean accuracy increased by 0.04 (p = 0.003), mean time to diagnosis reduced by 9.76 s (p < 0.001), and mean interrater agreement (Fleiss' Kappa) increased by 0.07 (p = 0.029). Conclusion: Our model stands proficient for labeling intracranial arteries using the largest CTA dataset. Crucially, it demonstrates clinical utility, helping prioritize the patients with high risks and ease clinical workload.

Artificial Intelligence in Symptomatic Carotid Plaque Detection: A Narrative Review

Identifying atherosclerotic disease is the mainstay for the correct diagnosis of the large artery atherosclerosis ischemic stroke subtype and for choosing the right therapeutic strategy in acute ischemic stroke. Classification into symptomatic and asymptomatic plaque and estimation of the cardiovascular risk are essential to select patients eligible for pharmacological and/or surgical therapy in order to prevent future cerebral ischemic events. The difficulties in a “vulnerability” definition and the methodical issues concerning its detectability and quantification are still subjects of debate. Non-invasive imaging studies commonly used to detect arterial plaque are computed tomographic angiography, magnetic resonance imaging, and ultrasound. Characterization of a carotid plaque type using the abovementioned imaging modalities represents the basis for carotid atherosclerosis management. Classification into symptomatic and asymptomatic plaque and estimation of the cardiovascular risk are essential to select patients eligible for pharmacological and/or surgical therapy in order to prevent future cerebral ischemic events. In this setting, artificial intelligence (AI) can offer suggestive solutions for tissue characterization and classification concerning carotid artery plaque imaging by analyzing complex data and using automated algorithms to obtain a final output. The aim of this review is to provide overall knowledge about the role of AI models applied to non-invasive imaging studies for the detection of symptomatic and vulnerable carotid plaques.

Edge Constraint and Location Mapping for Liver Tumor Segmentation from Nonenhanced Images

As there is no contrast enhancement, the liver tumor area in nonenhanced MRI exists with blurred edges and low contrast, which greatly affects the speed and accuracy of liver tumor diagnosis. As a result, precise segmentation of liver tumor from nonenhanced MRI has become an urgent and challenging task. In this paper, we propose an edge constraint and localization mapping segmentation model (ECLMS) to accurately segment liver tumor from nonenhanced MRI. It consists of two parts: localization network and dual-branch segmentation network. We build the localization network, which generates prior coarse masks to provide position mapping for the segmentation network. This part enhances the ability of the model to localize liver tumor in nonenhanced images. We design a dual-branch segmentation network, where the main decoding branch focuses on the feature representation in the core region of the tumor and the edge decoding branch concentrates on capturing the edge information of the tumor. To improve the ability of the model for capturing detailed features, sSE blocks and dense upward connections are introduced into it. We design the bottleneck multiscale module to construct multiscale feature representations using kernels of different sizes while integrating the location mapping of tumor. The ECLMS model is evaluated on a private nonenhanced MRI dataset that comprises 215 different subjects. The model achieves the best Dice coefficient, precision, and accuracy of 90.23%, 92.25%, and 92.39%, correspondingly. The effectiveness of our model is demonstrated by experiment results, and our model reaches superior results in the segmentation task of nonenhanced liver tumor compared to existing segmentation methods.

Diagnostic value of artificial intelligence automatic detection systems for breast BI-RADS 4 nodules

BACKGROUND

The incidence rate of breast cancer has exceeded that of lung cancer, and it has become the most malignant type of cancer in the world. BI-RADS 4 breast nodules have a wide range of malignant risks and are associated with challenging clinical decision-making.

AIM

To explore the diagnostic value of artificial intelligence (AI) automatic detection systems for BI-RADS 4 breast nodules and to assess whether conventional ultrasound BI-RADS classification with AI automatic detection systems can reduce the probability of BI-RADS 4 biopsy.

METHODS

A total of 107 BI-RADS breast nodules confirmed by pathology were selected between June 2019 and July 2020 at Hwa Mei Hospital, University of Chinese Academy of Sciences. These nodules were classified by ultrasound doctors and the AI-SONIC breast system. The diagnostic values of conventional ultrasound, the AI automatic detection system, conventional ultrasound combined with the AI automatic detection system and adjusted BI-RADS classification diagnosis were statistically analyzed.

RESULTS

Among the 107 breast nodules, 61 were benign (57.01%), and 46 were malignant (42.99%). The pathology results were considered the gold standard; furthermore, the sensitivity, specificity, accuracy, Youden index, and positive and negative predictive values were 84.78%, 67.21%, 74.77%, 0.5199, 66.10% and 85.42% for conventional ultrasound BI-RADS classification diagnosis, 86.96%, 75.41%, 80.37%, 0.6237, 72.73%, and 88.46% for automatic AI detection, 80.43%, 90.16%, 85.98%, 0.7059, 86.05%, and 85.94% for conventional ultrasound BI-RADS classification with automatic AI detection and 93.48%, 67.21%, 78.50%, 0.6069, 68.25%, and 93.18% for adjusted BI-RADS classification, respectively. The biopsy rate, cancer detection rate and malignancy risk were 100%, 42.99% and 0% and 67.29%, 61.11%, and 1.87% before and after BI-RADS adjustment, respectively.

CONCLUSION

Automatic AI detection has high accuracy in determining benign and malignant BI-RADS 4 breast nodules. Conventional ultrasound BI-RADS classification combined with AI automatic detection can reduce the biopsy rate of BI-RADS 4 breast nodules.