Automated unruptured cerebral aneurysms detection in TOF MR angiography images using dual-channel SE-3D UNet: a multi-center research

Global trend in research of intracranial aneurysm management with artificial intelligence technology: a bibliometric analysis

“…Studies such as those conducted by Lee et al ( 36 ) and Jin et al ( 37 ) have ventured into developing new deep-learning algorithms to overcome these challenges. Moreover, there has been a shift in the focus of research, with an increasing emphasis on time-of-flight MR angiography (TOF-MRA) as a primary non-invasive screening method for IAs, in contrast to the previous concentration on digital subtraction angiography (DSA) ( 38 ). Additionally, some studies have extended the application of AI to encompass both diagnosis and rupture prediction, exemplified by the work of Hentschke et al ( 39 ).…”

Advances in research and application of artificial intelligence and radiomic predictive models based on intracranial aneurysm images

“…This suggests that the architectural structure of different CNN models could impact detection and segmentation capabilities. Chen et al ( 20 ) developed an aneurysm detection model based on a dual-channel SE-3D UNet, retrospectively collecting 1,096 TOF-MRA images of unruptured intracranial aneurysms. The model, which divided the dataset into training and validation sets chronologically, outperformed the basic SE-3D UNet, increasing patient-level sensitivity by 15.79% and reducing false positives by 4.1%.…”

“…This suggests that the architectural structure of different CNN models could impact detection and segmentation capabilities. Chen et al (20) developed an aneurysm detection model based on a dual-channel SE-3D UNet, retrospectively collecting 1,096 TOF-MRA images of unruptured Despite promising results in automated detection studies, current capabilities still show significant room for improvement, especially in detecting small and very small aneurysms (22,23). This is related to the subtle imaging features of tiny aneurysms, which are easily overlooked or confused with normal physiological The study showed that the AI model achieved high diagnostic accuracy in the external validation set (sensitivity of 98.8%, specificity of 81.2%, and negative predictive value of 99.8%), surpassing the performance of radiologists, with an error rate of 0.5% in prospective validation.…”

A multi-center performance assessment for automated histopathological classification and grading of glioma using whole slide images

“…First, limited generalizability and single-center training may result in a model that is overly optimized for the specific characteristics of the training dataset. 16 , 17 As a result, the model may not generalize well to the different populations, imaging protocols, or equipment variations that exist in other institutions. This lack of generalizability could lead to poor performance when applied to new and diverse datasets.…”