Interpretable CNN for ischemic stroke subtype classification with active model adaptation.
Ontology highlight
ABSTRACT: Background
TOAST subtype classification is important for diagnosis and research of ischemic stroke. Limited by experience of neurologist and time-consuming manual adjudication, it is a big challenge to finish TOAST classification effectively. We propose a novel active deep learning architecture to classify TOAST.Methods
To simulate the diagnosis process of neurologists, we drop the valueless features by XGB algorithm and rank the remaining ones. Utilizing active learning framework, we propose a novel causal CNN, in which it combines with a mixed active selection criterion to optimize the uncertainty of samples adaptively. Meanwhile, KL-focal loss derived from the enhancement of Focal loss by KL regularization is introduced to accelerate the iterative fine-tuning of the model.Results
To evaluate the proposed method, we construct a dataset which consists of totally 2310 patients. In a series of sequential experiments, we verify the effectiveness of each contribution by different evaluation metrics. Experimental results show that the proposed method achieves competitive results on each evaluation metric. In this task, the improvement of AUC is the most obvious, reaching 77.4.Conclusions
We construct a backbone causal CNN to simulate the neurologist process of that could enhance the internal interpretability. The research on clinical data also indicates the potential application value of this model in stroke medicine. Future work we would consider various data types and more comprehensive patient types to achieve fully automated subtype classification.
SUBMITTER: Zhang S
PROVIDER: S-EPMC8729146 | biostudies-literature |
REPOSITORIES: biostudies-literature
Similar Datasets
Project description:BACKGROUND AND PURPOSE:In order to improve inter-rater reliability and minimize diagnosis of undetermined etiology for stroke subtype classification, using a stroke registry, we developed and implemented a magnetic resonance imaging (MRI)-based algorithm for acute ischemic stroke subtype classification (MAGIC). METHODS:We enrolled patients who experienced an acute ischemic stroke, were hospitalized in the 14 participating centers within 7 days of onset, and had relevant lesions on MR-diffusion weighted imaging (DWI). MAGIC was designed to reflect recent advances in stroke imaging and thrombolytic therapy. The inter-rater reliability was compared with and without MAGIC to classify the Trial of Org 10172 in Acute Stroke Treatment (TOAST) of each stroke patient. MAGIC was then applied to all stroke patients hospitalized since July 2011, and information about stroke subtypes, other clinical characteristics, and stroke recurrence was collected via a web-based registry database. RESULTS:The overall intra-class correlation coefficient (ICC) value was 0.43 (95% CI, 0.31-0.57) for MAGIC and 0.28 (95% CI, 0.18-0.42) for TOAST. Large artery atherosclerosis (LAA) was the most common cause of acute ischemic stroke (38.3%), followed by cardioembolism (CE, 22.8%), undetermined cause (UD, 22.2%), and small-vessel occlusion (SVO, 14.6%). One-year stroke recurrence rates were the highest for two or more UDs (11.80%), followed by LAA (7.30%), CE (5.60%), and SVO (2.50%). CONCLUSIONS:Despite several limitations, this study shows that the MAGIC system is feasible and may be helpful to classify stroke subtype in the clinic.
| S-EPMC4200592 | biostudies-literature
Project description:miRNAs (or microRNAs) are small, endogenous, and noncoding RNAs construct of about 22 nucleotides. Cumulative evidence from biological experiments shows that miRNAs play a fundamental and important role in various biological processes. Therefore, the classification of miRNA is a critical problem in computational biology. Due to the short length of mature miRNAs, many researchers are working on precursor miRNAs (pre-miRNAs) with longer sequences and more structural features. Pre-miRNAs can be divided into two groups as mirtrons and canonical miRNAs in terms of biogenesis differences. Compared to mirtrons, canonical miRNAs are more conserved and easier to be identified. Many existing pre-miRNA classification methods rely on manual feature extraction. Moreover, these methods focus on either sequential structure or spatial structure of pre-miRNAs. To overcome the limitations of previous models, we propose a nucleotide-level hybrid deep learning method based on a CNN and LSTM network together. The prediction resulted in 0.943 (%95 CI ± 0.014) accuracy, 0.935 (%95 CI ± 0.016) sensitivity, 0.948 (%95 CI ± 0.029) specificity, 0.925 (%95 CI ± 0.016) F1 Score and 0.880 (%95 CI ± 0.028) Matthews Correlation Coefficient. When compared to the closest results, our proposed method revealed the best results for Acc., F1 Score, MCC. These were 2.51%, 1.00%, and 2.43% higher than the closest ones, respectively. The mean of sensitivity ranked first like Linear Discriminant Analysis. The results indicate that the hybrid CNN and LSTM networks can be employed to achieve better performance for pre-miRNA classification. In future work, we study on investigation of new classification models that deliver better performance in terms of all the evaluation criteria.
| S-EPMC8266811 | biostudies-literature
Project description:BACKGROUND AND PURPOSE:Most cryptogenic strokes are thought to have an embolic source. We sought to determine whether cryptogenic strokes are associated with visceral infarcts, which are usually embolic. METHODS:Among patients prospectively enrolled in CAESAR (Cornell Acute Stroke Academic Registry), we selected those with a contrast-enhanced abdominal computed tomographic scan within 1 year of admission. Our exposure variable was adjudicated stroke subtype per the Trial of ORG 10172 in Acute Stroke Treatment classification. Our outcome was renal or splenic infarction as assessed by a single radiologist blinded to stroke subtype. We used Fisher exact test and multiple logistic regression to compare the prevalence of visceral infarcts among cardioembolic strokes, strokes of undetermined etiology, and noncardioembolic strokes (large- or small-vessel strokes). RESULTS:Among 227 patients with ischemic stroke and a contrast-enhanced abdominal computed tomographic scan, 59 had a visceral infarct (35 renal and 27 splenic). The prevalence of visceral infarction was significantly different among cardioembolic strokes (34.2%; 95% confidence interval [CI], 23.7%-44.6%), strokes of undetermined etiology (23.9%; 95% CI, 15.0%-32.8%), and strokes from large-artery atherosclerosis or small-vessel occlusion (12.5%; 95% CI, 1.8%-23.2%; P=0.03). In multiple logistic regression models adjusted for demographics and vascular comorbidities, we found significant associations with visceral infarction for both cardioembolic stroke (odds ratio, 3.5; 95% CI, 1.2-9.9) and stroke of undetermined source (odds ratio, 3.3; 95% CI, 1.1-10.5) as compared with noncardioembolic stroke. CONCLUSIONS:The prevalence of visceral infarction differed significantly across ischemic stroke subtypes. Cardioembolic and cryptogenic strokes were associated with a higher prevalence of visceral infarcts than noncardioembolic strokes.
| S-EPMC5828975 | biostudies-literature
Project description:Systemic cancer and ischemic stroke are common conditions and two of the most frequent causes of death among the elderly. The association between cancer and stroke has been reported worldwide. Stroke causes severe disability for cancer patients, while cancer increases the risk of stroke. Moreover, cancer-related stroke is expected to increase due to advances in cancer treatment and an aging population worldwide. Because cancer and stroke share risk factors (such as smoking and obesity) and treatment of cancer can increase the risk of stroke (e.g., accelerated atherosclerosis after radiation therapy), cancer may accelerate conventional stroke mechanisms (i.e., atherosclerosis, small vessel disease, and cardiac thrombus). In addition, active cancer and chemotherapy may enhance thrombin generation causing stroke related to coagulopathy. Patients with stroke due to cancer-related coagulopathy showed the characteristics findings of etiologic work ups, D-dimer levels, and infarct patterns. In this review, we summarized the frequency of cancer-related stroke among patients with ischemic stroke, mechanisms of stroke with in cancer patients, and evaluation and treatment of cancer-related stroke. We discussed the possibility of cancer-related stroke as a stroke subtype, and presented the most recent discoveries in the pathomechanisms and treatment of stroke due to cancer-related coagulopathy.
| S-EPMC7005348 | biostudies-literature
Project description:Inference of etiology from lesion pattern in acute magnetic resonance imaging is valuable for management and prognosis of acute stroke patients. This study aims to assess the value of three-dimensional geometrical lesion-shape descriptors for stroke-subtype classification, specifically regarding stroke of cardioembolic origin.Stroke Etiology was classified according to ASCOD in retrospectively selected patients with acute stroke. Lesions were segmented on diffusion-weighed datasets, and descriptors of lesion shape quantified: surface area, sphericity, bounding box volume, and ratio between bounding box and lesion volume. Morphological measures were compared between stroke subtypes classified by ASCOD and between patients with embolic stroke of cardiac and non-cardiac source.150 patients (mean age 77 years; 95% CI, 65-80 years; median NIHSS 6, range 0-22) were included. Group comparison of lesion shape measures demonstrated that lesions caused by small-vessel disease were smaller and more spherical compared to other stroke subtypes. No significant differences of morphological measures were detected between patients with cardioembolic and non-cardioembolic stroke.Stroke lesions caused by small vessel disease can be distinguished from other stroke lesions based on distinctive morphological properties. However, within the group of embolic strokes, etiology could not be inferred from the morphology measures studied in our analysis.
| S-EPMC5720627 | biostudies-literature
Project description:Background:Machine learning strategies are prominent tools for data analysis. Especially in life sciences, they have become increasingly important to handle the growing datasets collected by the scientific community. Meanwhile, algorithms improve in performance, but also gain complexity, and tend to neglect interpretability and comprehensiveness of the resulting models. Results:Generalized Matrix Learning Vector Quantization (GMLVQ) is a supervised, prototype-based machine learning method and provides comprehensive visualization capabilities not present in other classifiers which allow for a fine-grained interpretation of the data. In contrast to commonly used machine learning strategies, GMLVQ is well-suited for imbalanced classification problems which are frequent in life sciences. We present a Weka plug-in implementing GMLVQ. The feasibility of GMLVQ is demonstrated on a dataset of Early Folding Residues (EFR) that have been shown to initiate and guide the protein folding process. Using 27 features, an area under the receiver operating characteristic of 76.6% was achieved which is comparable to other state-of-the-art classifiers. The obtained model is accessible at https://biosciences.hs-mittweida.de/efpred/. Conclusions:The application on EFR prediction demonstrates how an easy interpretation of classification models can promote the comprehension of biological mechanisms. The results shed light on the special features of EFR which were reported as most influential for the classification: EFR are embedded in ordered secondary structure elements and they participate in networks of hydrophobic residues. Visualization capabilities of GMLVQ are presented as we demonstrate how to interpret the results.
| S-EPMC6321665 | biostudies-literature
Project description:Background and Purpose- Transthoracic echocardiography (TTE) is widely used in the ischemic stroke setting. In this study, we aim to investigate the yield of TTE in patients with ischemic stroke and known subtype and whether the admission troponin level improves the yield of TTE. Methods- Data were abstracted from a single-center prospective ischemic stroke database for 18 months and included all patients with ischemic stroke whose etiologic subtype could be obtained without the need of TTE. Unadjusted and adjusted regression models were built to determine whether positive cardiac troponin levels (?0.1 ng/mL) improve the yield of TTE, adjusting for demographic and clinical characteristics. Results- We identified 578 patients who met the inclusion criteria. TTE changed clinical management in 64 patients (11.1%), but intracardiac thrombus was detected in only 4 patients (0.7%). In multivariable models, there was an association between TTE changing management and positive serum troponin level (adjusted odds ratio, 4.26; 95% CI, 2.17-8.34; P<0.001). Conclusions- In patients with ischemic stroke, TTE might lead to a change in clinical management in ?1 of 10 patients with known stroke subtype before TTE but changed acute treatment decisions in <1 percent of patients. Serum troponin levels improved the yield of TTE in these patients.
| S-EPMC6209115 | biostudies-literature
Project description:Microglia are key regulators of inflammatory response after stroke and brain injury. Here we profiled the microglia transcriptome isolated from a spontaneously hypertensive rat model of focal cerebral ischemia. We identified an extensive and persistent upregulation of anti-inflammatory M2-like patterns after stroke and a mild up-regulation of pro-inflammatory M1-like patterns at later stage. We also found that younger brains showed larger microglial response than middle aged brains. Moreover, beyond the standard M1/M2 dichotomy, a wide spectrum of novel microglial polarization states was activated in response to stroke, particularly the phenotypes related to Tlr2 and dietary fatty acids stimulation.
2020-04-24 | GSE148350 | GEO
Project description:The implementation of uniform nomenclature and classification in adult arterial ischemic stroke (AIS) has been critical for defining outcomes and recurrence risks according to etiology and in developing risk-stratified treatments. In contrast, current classification and nomenclature in childhood AIS are often overlapping or contradictory. Our purpose was to develop a comprehensive consensus-based classification system for childhood AIS.Using a modified-Delphi method, members of the International Pediatric Stroke Study (IPSS) developed the Childhood AIS Standardized Classification And Diagnostic Evaluation (CASCADE) criteria. Two groups of pediatric stroke specialists from the IPSS classified 7 test cases using 2 methods each: (1) classification typical of the individual clinician's current clinical practice; and (2) classification based on the CASCADE criteria. Group 1 underwent in-person training in the utilization of the CASCADE criteria. Group 2 classified the same cases via an online survey, including definitions but without training. Inter-rater reliability (IRR) was assessed via multi-rater unweighted ?-statistic.In Group 1 (with training), IRR was improved using CASCADE criteria (?=0.78, 95% CI=[0.49, 0.94]), compared with typical clinical practice (?=0.40, 95% CI=[0.11, 0.60]). In Group 2 (without training), IRR was lower than among trained raters (?=0.61, 95% CI=[0.29, 0.77]), but higher than current practice (?=0.23, 95% CI=[0.03, 0.36]).A new, consensus-based classification system for childhood AIS, the CASCADE criteria, can be used to classify cases with good IRR. These preliminary findings suggest that the CASCADE criteria may be particularity useful in the setting of prospective multicenter studies in childhood-onset AIS, where standardized training of investigators is feasible.
| S-EPMC3312781 | biostudies-literature
Project description:Classification of benign and malignant in lung nodules using chest CT images is a key step in the diagnosis of early-stage lung cancer, as well as an effective way to improve the patients' survival rate. However, due to the diversity of lung nodules and the visual similarity of lung nodules to their surrounding tissues, it is difficult to construct a robust classification model with conventional deep learning-based diagnostic methods. To address this problem, we propose a multi-model ensemble learning architecture based on 3D convolutional neural network (MMEL-3DCNN). This approach incorporates three key ideas: (1) Constructed multi-model network architecture can be well adapted to the heterogeneity of lung nodules. (2) The input that concatenated of the intensity image corresponding to the nodule mask, the original image, and the enhanced image corresponding to which can help training model to extract advanced feature with more discriminative capacity. (3) Select the corresponding model to different nodule size dynamically for prediction, which can improve the generalization ability of the model effectively. In addition, ensemble learning is applied in this paper to further improve the robustness of the nodule classification model. The proposed method has been experimentally verified on the public dataset, LIDC-IDRI. The experimental results show that the proposed MMEL-3DCNN architecture can obtain satisfactory classification results.
| S-EPMC7649841 | biostudies-literature