SINO-CT-Fusion-Net: A Lightweight Deep Learning Framework for Detection and Classification of Intracranial Hemorrhages

Performance of an artificial intelligence tool with real-time clinical workflow integration - Detection of intracranial hemorrhage and pulmonary embolism.

Purpose: Recently developed machine-learning algorithms have demonstrated strong performance in the detection of intracranial hemorrhage (ICH) and large vessel occlusion (LVO). However, their generalizability is often limited by geographic bias of studies. The aim of this study was to validate a commercially available deep learning-based tool in the detection of both ICH and LVO across multiple hospital sites and vendors throughout the U.S.Materials and Methods: This was a retrospective and multicenter study using anonymized data from two institutions. Eight hundred fourteen non-contrast CT cases and 378 CT angiography cases were analyzed to evaluate ICH and LVO, respectively. The tool's ability to detect and quantify ICH, LVO, and their various subtypes was assessed among multiple CT vendors and hospitals across the United States. Ground truth was based off imaging interpretations from two board-certified neuroradiologists.Results: There were 255 positive and 559 negative ICH cases. Accuracy was 95.6%, sensitivity was 91.4%, and specificity was 97.5% for the ICH tool. ICH was further stratified into the following subtypes: intraparenchymal, intraventricular, epidural/subdural, and subarachnoid with true positive rates of 92.9, 100, 94.3, and 89.9%, respectively. ICH true positive rates by volume [small (<5 mL), medium (5–25 mL), and large (>25 mL)] were 71.8, 100, and 100%, respectively. There were 156 positive and 222 negative LVO cases. The LVO tool demonstrated an accuracy of 98.1%, sensitivity of 98.1%, and specificity of 98.2%. A subset of 55 randomly selected cases were also assessed for LVO detection at various sites, including the distal internal carotid artery, middle cerebral artery M1 segment, proximal middle cerebral artery M2 segment, and distal middle cerebral artery M2 segment with an accuracy of 97.0%, sensitivity of 94.3%, and specificity of 97.4%.Conclusion: Deep learning tools can be effective in the detection of both ICH and LVO across a wide variety of hospital systems. While some limitations were identified, specifically in the detection of small ICH and distal M2 occlusion, this study highlights a deep learning tool that can assist radiologists in the detection of emergent findings in a variety of practice settings.

SummaryIn this article, the detection and categorization of acute intracranial hemorrhage (ICH) subtypes using a multilayer DenseNet‐ResNet architecture with improved random forest classifier (IRF) is proposed to detect the subtypes of intracerebral hemorrhage with high accuracy and less computational time. Here, the brain CT images are taken from the physionet repository publicly dataset. Then the images are preprocessed to eliminate the unwanted noises. After that, the image features are extracted by using multilayer densely connected convolutional network (DenseNet) combined with residual network (ResNet) architecture with multiple convolutional layers. The subtypes are epidural hemorrhage (EDH), subarachnoid hemorrhage (SAH), intraparenchymal hemorrhage (IPH), subdural hemorrhage (SDH), intraventricular hemorrhage (IVH) are classified by using an IRF classifier with high accuracy. The simulation process is carried out in MATLAB site. The proposed multilayer‐DenseNet‐ResNet‐IRF attains higher accuracy 23.44%, 31.93%, 42.83%, 41.9% is compared with the existing methods, such as deep learning algorithm for automatic detection and classification of acute intracranial hemorrhages in head CT scans (ICH‐DC‐2D‐CNN), fusion‐based deep learning along nature‐inspired algorithm for the diagnosis of intracerebral hemorrhage (ICH‐DC‐FSVM), and detection of intracranial hemorrhage on CT scan images using convolutional neural network (ICH‐DC‐CNN) and double fully convolutional networks (FCNs), respectively.

Purpose To apply conformal prediction to a deep learning (DL) model for intracranial hemorrhage (ICH) detection and evaluate model performance in detection as well as model accuracy in identifying challenging cases. Materials and Methods This was a retrospective (November-December 2017) study of 491 noncontrast head CT volumes from the CQ500 dataset, in which three senior radiologists annotated sections containing ICH. The dataset was split into definite and challenging (uncertain) subsets, in which challenging images were defined as those in which there was disagreement among readers. A DL model was trained on patients from the definite data (training dataset) to perform ICH localization and classification into five classes. To develop an uncertainty-aware DL model, 1546 sections of the definite data (calibration dataset) were used for Mondrian conformal prediction (MCP). The uncertainty-aware DL model was tested on 8401 definite and challenging sections to assess its ability to identify challenging sections. The difference in predictive performance (P value) and ability to identify challenging sections (accuracy) were reported. Results The study included 146 patients (mean age, 45.7 years ± 9.9 [SD]; 76 [52.1%] men, 70 [47.9%] women). After the MCP procedure, the model achieved an F1 score of 0.919 for localization and classification. Additionally, it correctly identified patients with challenging cases with 95.3% (143 of 150) accuracy. It did not incorrectly label any definite sections as challenging. Conclusion The uncertainty-aware MCP-augmented DL model achieved high performance in ICH detection and high accuracy in identifying challenging sections, suggesting its usefulness in automated ICH detection and potential to increase trustworthiness of DL models in radiology. Keywords: CT, Head and Neck, Brain, Brain Stem, Hemorrhage, Feature Detection, Diagnosis, Supervised Learning Supplemental material is available for this article. © RSNA, 2025 See also commentary by Ngum and Filippi in this issue.

Background and purposeMultiple attempts at intracranial hemorrhage (ICH) detection using deep-learning techniques have been plagued by clinical failures. We aimed to compare the performance of a deep-learning algorithm for ICH detection trained on strongly and weakly annotated datasets, and to assess whether a weighted ensemble model that integrates separate models trained using datasets with different ICH improves performance.MethodsWe used brain CT scans from the Radiological Society of North America (27,861 CT scans, 3,528 ICHs) and AI-Hub (53,045 CT scans, 7,013 ICHs) for training. DenseNet121, InceptionResNetV2, MobileNetV2, and VGG19 were trained on strongly and weakly annotated datasets and compared using independent external test datasets. We then developed a weighted ensemble model combining separate models trained on all ICH, subdural hemorrhage (SDH), subarachnoid hemorrhage (SAH), and small-lesion ICH cases. The final weighted ensemble model was compared to four well-known deep-learning models. After external testing, six neurologists reviewed 91 ICH cases difficult for AI and humans.ResultsInceptionResNetV2, MobileNetV2, and VGG19 models outperformed when trained on strongly annotated datasets. A weighted ensemble model combining models trained on SDH, SAH, and small-lesion ICH had a higher AUC, compared with a model trained on all ICH cases only. This model outperformed four deep-learning models (AUC [95% C.I.]: Ensemble model, 0.953[0.938–0.965]; InceptionResNetV2, 0.852[0.828–0.873]; DenseNet121, 0.875[0.852–0.895]; VGG19, 0.796[0.770–0.821]; MobileNetV2, 0.650[0.620–0.680]; p < 0.0001). In addition, the case review showed that a better understanding and management of difficult cases may facilitate clinical use of ICH detection algorithms.ConclusionWe propose a weighted ensemble model for ICH detection, trained on large-scale, strongly annotated CT scans, as no model can capture all aspects of complex tasks.

To evaluate the post-deployment performance of an artificial intelligence (AI) system (Aidoc) for intracranial hemorrhage (ICH) detection and assess the utility of ChatGPT-4 Turbo for automated AI monitoring. This retrospective study evaluated 332,809 head CT examinations from 37 radiology practices across the United States (December 2023-May 2024). Of these, 13,569 cases were flagged as positive for ICH by the Aidoc AI system. A HIPAA (Health Insurance Portability and Accountability Act) -compliant version of ChatGPT-4 Turbo was used to extract data from radiology reports. Ground truth was established through radiologists' review of 200 randomly selected cases. Performance metrics were calculated for ChatGPT, Aidoc and radiologists. ChatGPT-4 Turbo demonstrated high diagnostic accuracy in identifying intracranial hemorrhage (ICH) from radiology reports, with a positive predictive value of 1 and a negative predictive value of 0.988 (AUC:0.996). Aidoc's false positive classifications were influenced by scanner manufacturer, midline shift, mass effect, artifacts, and neurologic symptoms. Multivariate analysis identified Philips scanners (OR: 6.97, p=0.003) and artifacts (OR: 3.79, p=0.029) as significant contributors to false positives, while midline shift (OR: 0.08, p=0.021) and mass effect (OR: 0.18, p=0.021) were associated with a reduced false positive rate. Aidoc-assisted radiologists achieved a sensitivity of 0.936 and a specificity of 1. This study underscores the importance of continuous performance monitoring for AI systems in clinical practice. The integration of LLMs offers a scalable solution for evaluating AI performance, ensuring reliable deployment and enhancing diagnostic workflows.

Intracranial hemorrhage (ICH) detection is the primary task for the patients suffering from neurological disturbances and head injury. This paper presents a segmentation technique that combines the features of fuzzy c-mean (FCM) clustering and region-based active contour method. In the suggested method, the fuzzy membership degree from FCM clustering is first used to initialize the active contour, which propagates for the detection of the desired object. In addition to active contour initialization, the fuzzy clustering is also used to estimate the contour propagation controlling parameters. The level set function as used by active contour in the proposed method does not need re-initialization process; thus, it fastens the convergent speed of the contour propagation. The efficacy of the suggested method is demonstrated on a dataset of 20 brain computed tomography (CT) images suffered with ICH. Experimental results show that the proposed method has advantages in accuracy in comparison with standard region growing method and FCM for the detection of hemorrhage from brain CT images.

Focus issue: Artificial intelligence in medical physics.

Intracranial haemorrhage is a bleeding that occurs in brain which needs immediate medical attention and intensive medical care. The objective of this work is early detection of intracranial haemorrhage through automated model using DenseNets. DenseNets are used for processing MRI images and for detection of intracranial haemorrhage and its different variants. MRI scanned images samples are collected from a nearby neurology super speciality hospital. Segmentation of images is done through DenseNets which are also called deep connected convolution networks. Based on the image segments, the variant of intracranial haemorrhage is predicted. DenseNets layers are very narrow and as they add small set of feature maps and performs better when compared to the detection of the intracranial haemorrhage using convolution neural network (Juan et al in Proceedings of 4th congress on robotics and neuro science (2019), []). The accuracy of the proposed method is 91% achieved through the gradient from loss function which has access to each and every layer.

The assessment of the presence of intracranial hemorrhage is a crucial step in the work-up of patients requiring emergency care. Fast and accurate detection of intracranial hemorrhage can aid treating physicians by not only expediting and guiding diagnosis, but also supporting choices for secondary imaging, treatment and intervention. However, the automatic detection of intracranial hemorrhage is complicated by the variation in appearance on non-contrast CT images as a result of differences in etiology and location. We propose a method using a convolutional neural network (CNN) for the automatic detection of intracranial hemorrhage. The method is trained on a dataset comprised of cerebral CT studies for which the presence of hemorrhage has been labeled for each axial slice. A separate test dataset of 20 images is used for quantitative evaluation and shows a sensitivity of 0.87, specificity of 0.97 and accuracy of 0.95. The average processing time for a single three-dimensional (3D) CT volume was 2.7 seconds. The proposed method is capable of fast and automated detection of intracranial hemorrhages in non-contrast CT without being limited to a specific subtype of pathology.

Missed detection of intracranial hemorrhage in Head CT scans has significantly impacted patient morbidity and mortality. Early detection of intracranial hemorrhage enables patients to receive appropriate treatment which resulted in a better outcome. Some doctors have limited experience in interpreting the CT scan hence increasing the probability to miss the hemorrhage. The main objective of this study is to develop an algorithm model capable of detecting intracranial hemorrhage in a head CT scan. We are using deep learning from a convolutional neural network (CNN) to produce this algorithm module. This was a cross-sectional study using secondary data, in which 200 data were collected from public datasets. All of the samples have been anonymized into secondary data. The algorithm model is trained using deep learning via a Jupyter Notebook platform. To analyze the algorithm model performance, a confusion matrix was used to measure the accuracy, sensitivity, specificity, precision, and F1 score. This study showed that from 200 training data, 95 samples were true positive, 95 samples were true negative, 7 samples were false positive, and 3 samples were false negative. This algorithm model shows high sensitivity (0.9694), high specificity (0.9314), high precision (0.9314), and high accuracy (0.9500) with an F1 score of 0.9500. This study has proven that deep learning by using CNN enables us to create an accurate classifier that can differentiate between head CT scan with intracranial hemorrhage and without hemorrhage. The main objective of this study is to develop an algorithm model capable of detecting intracranial hemorrhage in a head CT scan. We are using deep learning from a convolutional neural network (CNN) to produce this algorithm module. This was a cross-sectional study using secondary data, in which 200 data were collected from public datasets. This dataset is owned by Abdul Kader Helwan, an academic staff at Al-Manar University of Tripoli, Lebanon. Permission to use the dataset for this research was officially obtained from the owner. All of samples have been anonymized into secondary data. The data is divided into train, validation, and test samples. The algorithm model is trained using deep learning via a Jupyter Notebook platform. To analyze the algorithm model performance, confusion matrix was used to measure the accuracy, sensitivity, specificity, precision, and F1 score. This study showed that from 200 training data, 95 samples were true positive, 95 samples were true negative, 7 samples were false positive, and 3 samples were false negative. This algorithm model shows high sensitivity (0.9694), high specificity (0.9314), high precision (0.9314), and high accuracy (0.9500) with F1 score of 0.9500. Hence, this study has proven that deep learning by using CNN enables us to create an accurate classifier that can differentiate between head CT scan with intracranial hemorrhage and without hemorrhage.

Purpose:To design a dedicated x‐ray cone‐beam CT (CBCT) system suitable to deployment at the point‐of‐care and offering reliable detection of acute intracranial hemorrhage (ICH), traumatic brain injury (TBI), stroke, and other head and neck injuries.Methods:A comprehensive task‐based image quality model was developed to guide system design and optimization of a prototype head scanner suitable to imaging of acute TBI and ICH. Previously reported models were expanded to include the effects of x‐ray scatter correction necessary for detection of low contrast ICH and the contribution of bit depth (digitization noise) to imaging performance. Task‐based detectablity index provided the objective function for optimization of system geometry, x‐ray source, detector type, anti‐scatter grid, and technique at 10–25 mGy dose. Optimal characteristics were experimentally validated using a custom head phantom with 50 HU contrast ICH inserts imaged on a CBCT imaging bench allowing variation of system geometry, focal spot size, detector, grid selection, and x‐ray technique.Results:The model guided selection of system geometry with a nominal source‐detector distance 1100 mm and optimal magnification of 1.50. Focal spot size ∼0.6 mm was sufficient for spatial resolution requirements in ICH detection. Imaging at 90 kVp yielded the best tradeoff between noise and contrast. The model provided quantitation of tradeoffs between flat‐panel and CMOS detectors with respect to electronic noise, field of view, and readout speed required for imaging of ICH. An anti‐scatter grid was shown to provide modest benefit in conjunction with post‐acquisition scatter correction. Images of the head phantom demonstrate visualization of millimeter‐scale simulated ICH.Conclusions:Performance consistent with acute TBI and ICH detection is feasible with model‐based system design and robust artifact correction in a dedicated head CBCT system. Further improvements can be achieved with incorporation of model‐based iterative reconstruction techniques also within the scope of the task‐based optimization framework.David Foos and Xiaohui Wang are employees of Carestream Health

Machine learning (ML) holds great promise in transforming healthcare. While published studies have shown the utility of ML models in interpreting medical imaging examinations, these are often evaluated under laboratory settings. The importance of real world evaluation is best illustrated by case studies that have documented successes and failures in the translation of these models into clinical environments. A key prerequisite for the clinical adoption of these technologies is demonstrating generalizable ML model performance under real world circumstances. The purpose of this study was to demonstrate that ML model generalizability is achievable in medical imaging with the detection of intracranial hemorrhage (ICH) on non-contrast computed tomography (CT) scans serving as the use case. An ML model was trained using 21,784 scans from the RSNA Intracranial Hemorrhage CT dataset while generalizability was evaluated using an external validation dataset obtained from our busy trauma and neurosurgical center. This real world external validation dataset consisted of every unenhanced head CT scan (n = 5965) performed in our emergency department in 2019 without exclusion. The model demonstrated an AUC of 98.4%, sensitivity of 98.8%, and specificity of 98.0%, on the test dataset. On external validation, the model demonstrated an AUC of 95.4%, sensitivity of 91.3%, and specificity of 94.1%. Evaluating the ML model using a real world external validation dataset that is temporally and geographically distinct from the training dataset indicates that ML generalizability is achievable in medical imaging applications.

Introduction: Prompt detection of intracranial hemorrhage (ICH) on a non-contrast head CT (NCCT) is critical to initial patient triage. Several ICH detection and notification programs now exist, but none presently generate locations or segmentations of suspected ICH, and false-positive notifications are not easily identified. We investigated an enhanced ICH detection tool with heightened immunity to false-positive diagnosis. Methods: NCCT scans from 3 large databases were evaluated for the presence of an ICH (IPH, IVH, SAH or SDH) of &gt;0.4 ml by the automated RAPID ICH 3.0 software module and compared to a consensus detection from 3 neuroradiology experts. Scans were excluded for (1) severe CT acquisition artifacts, (2) prior neurosurgical procedures or (3) intravenous contrast administration. ICH detection accuracy, sensitivity, specificity, positive predictive value, negative predictive value, and positive and negative likelihood ratios by RAPID ICH 3 were determined. Results: A total of 790 studies were included. RAPID ICH 3.0 correctly identified 399/411 ICH-positive cases and 378/379 ICH-negative cases, resulting in a sensitivity of 98.08% and specificity 99.74%, positive predictive value 99.75%, and negative predictive value 96.92% for ICH detection. The positive and negative likelihood ratios for ICH detection were similarly favorable at 367.93 and 0.03 respectively. Mean processing time was &lt;40 seconds. Conclusions: In this large data set of nearly 800 patients, the RAPID ICH 3.0 automated software maintained a high sensitivity for detection of ICH, while essentially eliminating false positive identifications, leading to very high positive predictive value.

The need for accurate and timely detection of Intracranial hemorrhage (ICH) is of utmost importance to avoid untoward incidents that may even lead to death. Hence, this presented work leverages the ability of a pretrained deep convolutional neural network (CNN) for the detection of ICH in computed tomography (CT) brain images. Different frameworks have been analyzed for their effectiveness for the classification of CT brain images into hemorrhage or non-hemorrhage conditions. All these frameworks were investigated on the CQ500 dataset. Furthermore, an exclusive preprocessing pipeline was designed for both normal and ICH CT images. Firstly, a framework involving the pretrained deep CNN, AlexNet, has been exploited for both feature extraction and classification using the transfer learning method. Secondly, a modified AlexNet-Support vector machine (SVM) classifier is explored, and finally, a feature selection method, Principal Component Analysis (PCA), has been introduced in the AlexNet-SVM classifier model, and its efficacy is also explored. These models were trained and tested on two different sets of CT images, one containing the original images without preprocessing and another set consisting of preprocessed images. The modified AlexNet-SVM classifier has shown an improved performance in comparison to the other investigated frameworks and has achieved a classification accuracy of 99.86% and sensitivity and specificity of 0.9986 for the detection of ICH in the brain CT images. This research has given an overview of a simple and efficient framework for the classification of hemorrhage and non-hemorrhage images. The proposed simplified deep learning framework also manifests its ability as a screening tool to assist the radiological trainees in the accurate detection of ICH.