Validation of the NeuroImaging Radiological Interpretation System for Acute Traumatic Brain Injury:

Enhancing hospital course and outcome prediction in patients with traumatic brain injury: A machine learning study

PURPOSE

We aimed to use machine learning (ML) algorithms with clinical, lab, and imaging data as input to predict various outcomes in traumatic brain injury (TBI) patients.

METHODS

In this retrospective study, blood samples were analyzed for glial fibrillary acidic protein (GFAP) and ubiquitin C-terminal hydrolase L1 (UCH-L1). The non-contrast head CTs were reviewed by two neuroradiologists for TBI common data elements (CDE). Three outcomes were designed to predict: discharged or admitted for further management (prediction 1), deceased or not deceased (prediction 2), and admission only, prolonged stay, or neurosurgery performed (prediction 3). Five ML models were trained. SHapley Additive exPlanations (SHAP) analyses were used to assess the relative significance of variables.

RESULTS

Four hundred forty patients were used to predict predictions 1 and 2, while 271 patients were used in prediction 3. Due to Prediction 3's hospitalization requirement, deceased and discharged patients could not be utilized. The Random Forest model achieved an average accuracy of 1.00 for prediction 1 and an accuracy of 0.99 for prediction 2. The Random Forest model achieved a mean accuracy of 0.93 for prediction 3. Key features were extracranial injury, hemorrhage, UCH-L1 for prediction 1; The Glasgow Coma Scale, age, GFAP for prediction 2; and GFAP, subdural hemorrhage volume, and pneumocephalus for prediction 3, per SHAP analysis.

CONCLUSION

Combining clinical and laboratory parameters with non-contrast CT CDEs allowed our ML models to accurately predict the designed outcomes of TBI patients. GFAP and UCH-L1 were among the significant predictor variables, demonstrating the importance of these biomarkers.

Effect of antiplatelet and anticoagulant medication use on injury severity and mortality in patients with traumatic brain injury treated in the intensive care unit

BACKGROUND

Antiplatelet and anticoagulant medication are increasingly common and can increase the risks of morbidity and mortality in traumatic brain injury (TBI) patients. Our study aimed to quantify the association of antiplatelet or anticoagulant use in intensive care unit (ICU)-treated TBI patients with 1-year mortality and head CT findings.

METHOD

We conducted a retrospective, multicenter observational study using the Finnish Intensive Care Consortium database. We included adult TBI patients admitted to four university hospital ICUs during 2003-2013. The patients were followed up until the end of 2016. The national drug reimbursement database provided information on prescribed medication for our study. We used multivariable logistic regression models to assess the association between TBI severity, prescribed antiplatelet and anticoagulant medication, and their association with 1-year mortality.

RESULTS

Of 3031 patients, 128 (4%) had antiplatelet and 342 (11%) anticoagulant medication before their TBI. Clopidogrel (2%) and warfarin (9%) were the most common antiplatelets and anticoagulants. Three patients had direct oral anticoagulant (DOAC) medication. The median age was higher among antiplatelet/anticoagulant users than in non-users (70 years vs. 52 years, p < 0.001), and their head CT findings were more severe (median Helsinki CT score 3 vs. 2, p < 0.05). In multivariable analysis, antiplatelets (OR 1.62, 95% CI 1.02-2.58) and anticoagulants (OR 1.43, 95% CI 1.06-1.94) were independently associated with higher odds of 1-year mortality. In a sensitivity analysis including only patients over 70, antiplatelets (OR 2.28, 95% CI 1.16-4.22) and anticoagulants (1.50, 95% CI 0.97-2.32) were associated with an increased risk of 1-year mortality.

CONCLUSIONS

Both antiplatelet and anticoagulant use before TBI were risk factors in our study for 1-year mortality. Antiplatelet and anticoagulation medication users had a higher radiological intracranial injury burden than non-users defined by the Helsinki CT score. Further investigation on the effect of DOACs on mortality should be done in ICU-treated TBI patients.

Validation of a deep learning model for traumatic brain injury detection and NIRIS grading on non-contrast CT: a multi-reader study with promising results and opportunities for improvement

PURPOSE

This study aimed to assess and externally validate the performance of a deep learning (DL) model for the interpretation of non-contrast computed tomography (NCCT) scans of patients with suspicion of traumatic brain injury (TBI).

METHODS

This retrospective and multi-reader study included patients with TBI suspicion who were transported to the emergency department and underwent NCCT scans. Eight reviewers, with varying levels of training and experience (two neuroradiology attendings, two neuroradiology fellows, two neuroradiology residents, one neurosurgery attending, and one neurosurgery resident), independently evaluated NCCT head scans. The same scans were evaluated using the version 5.0 of the DL model icobrain tbi. The establishment of the ground truth involved a thorough assessment of all accessible clinical and laboratory data, as well as follow-up imaging studies, including NCCT and magnetic resonance imaging, as a consensus amongst the study reviewers. The outcomes of interest included neuroimaging radiological interpretation system (NIRIS) scores, the presence of midline shift, mass effect, hemorrhagic lesions, hydrocephalus, and severe hydrocephalus, as well as measurements of midline shift and volumes of hemorrhagic lesions. Comparisons using weighted Cohen's kappa coefficient were made. The McNemar test was used to compare the diagnostic performance. Bland-Altman plots were used to compare measurements.

RESULTS

One hundred patients were included, with the DL model successfully categorizing 77 scans. The median age for the total group was 48, with the omitted group having a median age of 44.5 and the included group having a median age of 48. The DL model demonstrated moderate agreement with the ground truth, trainees, and attendings. With the DL model's assistance, trainees' agreement with the ground truth improved. The DL model showed high specificity (0.88) and positive predictive value (0.96) in classifying NIRIS scores as 0-2 or 3-4. Trainees and attendings had the highest accuracy (0.95). The DL model's performance in classifying various TBI CT imaging common data elements was comparable to that of trainees and attendings. The average difference for the DL model in quantifying the volume of hemorrhagic lesions was 6.0 mL with a wide 95% confidence interval (CI) of - 68.32 to 80.22, and for midline shift, the average difference was 1.4 mm with a 95% CI of - 3.4 to 6.2.

CONCLUSION

While the DL model outperformed trainees in some aspects, attendings' assessments remained superior in most instances. Using the DL model as an assistive tool benefited trainees, improving their NIRIS score agreement with the ground truth. Although the DL model showed high potential in classifying some TBI CT imaging common data elements, further refinement and optimization are necessary to enhance its clinical utility.

Comparing Predictive Utility of Head Computed Tomography Scan-Based Scoring Systems for Traumatic Brain Injury: A Retrospective Study

This study compared the predictive utility of Marshall, Rotterdam, Stockholm, Helsinki, and NeuroImaging Radiological Interpretation System (NIRIS) scorings based on early non-contrast brain computed tomography (CT) scans in patients with traumatic brain injury (TBI). The area under a receiver operating characteristic curve (AUROC) was used to determine the predictive utility of scoring systems. Subgroup analyses were performed among patients with head AIS scores > 1. A total of 996 patients were included, of whom 786 (78.9%) were males. In-hospital mortality, ICU admission, neurosurgical intervention, and prolonged total hospital length of stay (THLOS) were recorded for 27 (2.7%), 207 (20.8%), 82 (8.2%), and 205 (20.6%) patients, respectively. For predicting in-hospital mortality, all scoring systems had AUROC point estimates above 0.9 and 0.75 among all included patients and patients with head AIS > 1, respectively, without any significant differences. The Marshall and NIRIS scoring systems had higher AUROCs for predicting ICU admission and neurosurgery than the other scoring systems. For predicting THLOS ≥ seven days, although the NIRIS and Marshall scoring systems seemed to have higher AUROC point estimates when all patients were analyzed, five scoring systems performed roughly the same in the head AIS > 1 subgroup.

Prediction of In-Hospital Outcomes in Patients with Traumatic Brain Injury Using Computed Tomographic Scoring Systems: A Comparison Between Marshall, Rotterdam, and Neuroimaging Radiological Interpretation Systems

OBJECTIVE

This study aimed to compare the prognostic value of Marshall, Rotterdam, and Neuroimaging Radiological Interpretation Systems (NIRIS) in predicting the in-hospital outcomes of patients with traumatic brain injury.

METHODS

We identified 250 patients with traumatic brain injury in a retrospective single-center cohort from 2019 to 2020. Computed tomography (CT) scans were reviewed by two radiologists and scored according to three CT scoring systems. One-month outcomes were evaluated, including hospitalization, intensive care unit admission, neurosurgical procedure, and mortality. Logistic regression analysis was performed to identify scoring systems and outcome relationships. The best cutoff value was calculated using the receiver operating characteristic curve model.

RESULTS

Eighteen patients (7.2%) died in the 1-month follow-up. The mean age and Glasgow Coma Scale of survivors differed significantly from nonsurvivors. Subarachnoid hemorrhage and compressed/absent cisterns were dead patients' most frequent CT findings. All three scoring systems had good discrimination power in mortality prediction (area under the receiver operating characteristic curve of the Marshall, Rotterdam, and NIRIS was 0.78, 0.86, and 0.84, respectively). Regarding outcome, three systems directly correlated with unfavorable outcome prediction.

CONCLUSIONS

The Marshall, Rotterdam, and NIRIS are good predictive models for mortality and outcome prediction, with slight superiority of the Rotterdam in mortality prediction and the Marshall in intensive care unit admission and neurosurgical procedures.

Accuracy of head computed tomography scoring systems in predicting outcomes for patients with moderate to severe traumatic brain injury: A ProTECT III ancillary study

BACKGROUND

Several types of head CT classification systems have been developed to prognosticate and stratify TBI patients.

OBJECTIVE

The purpose of our study was to compare the predictive value and accuracy of the different CT scoring systems, including the Marshall, Rotterdam, Stockholm, Helsinki, and NIRIS systems, to inform specific patient management actions, using the ProTECT III population of patients with moderate to severe acute traumatic brain injury (TBI).

METHODS

We used the data collected in the patients with moderate to severe (GCS score of 4-12) TBI enrolled in the ProTECT III clinical trial. ProTECT III was a NIH-funded, prospective, multicenter, randomized, double-blind, placebo-controlled clinical trial designed to determine the efficacy of early administration of IV progesterone. The CT scoring systems listed above were applied to the baseline CT scans obtained in the trial. We assessed the predictive accuracy of these scoring systems with respect to Glasgow Outcome Scale-Extended at 6 months, disability rating scale score, and mortality.

RESULTS

A total of 882 subjects were enrolled in ProTECT III. Worse scores for each head CT scoring systems were highly correlated with unfavorable outcome, disability outcome, and mortality. The NIRIS classification was more strongly correlated than the Stockholm and Rotterdam CT scores, followed by the Helsinki and Marshall CT classification. The highest correlation was observed between NIRIS and mortality (estimated odds ratios of 4.83).

CONCLUSION

All scores were highly associated with 6-month unfavorable, disability and mortality outcomes. NIRIS was also accurate in predicting TBI patients' management and disposition.

External validation of the NeuroImaging Radiological Interpretation System and Helsinki computed tomography score for mortality prediction in patients with traumatic brain injury treated in the intensive care unit: a Finnish intensive care consortium study

BACKGROUND

Admission computed tomography (CT) scoring systems can be used to objectively quantify the severity of traumatic brain injury (TBI) and aid in outcome prediction. We aimed to externally validate the NeuroImaging Radiological Interpretation System (NIRIS) and the Helsinki CT score. In addition, we compared the prognostic performance of the NIRIS and the Helsinki CT score to the Marshall CT classification and to a clinical model.

METHODS

We conducted a retrospective multicenter observational study using the Finnish Intensive Care Consortium database. We included adult TBI patients admitted in four university hospital ICUs during 2003-2013. We analyzed the CT scans using the NIRIS and the Helsinki CT score and compared the results to 6-month mortality as the primary outcome. In addition, we created a clinical model (age, Glasgow Coma Scale score, Simplified Acute Physiology Score II, presence of severe comorbidity) and combined clinical and CT models to see the added predictive impact of radiological data to conventional clinical information. We measured model performance using area under curve (AUC), Nagelkerke's R² statistics, and the integrated discrimination improvement (IDI).

RESULTS

A total of 3031 patients were included in the analysis. The 6-month mortality was 710 patients (23.4%). Of the CT models, the Helsinki CT displayed best discrimination (AUC 0.73 vs. 0.70 for NIRIS) and explanatory variation (Nagelkerke's R² 0.20 vs. 0.15). The clinical model displayed an AUC of 0.86 (95% CI 0.84-0.87). All CT models increased the AUC of the clinical model by + 0.01 to 0.87 (95% CI 0.85-0.88) and the IDI by 0.01-0.03.

CONCLUSION

In patients with TBI treated in the ICU, the Helsinki CT score outperformed the NIRIS for 6-month mortality prediction. In isolation, CT models offered only moderate accuracy for outcome prediction and clinical variables outweighing the CT-based predictors in terms of predictive performance.

ACR Appropriateness Criteria® Head Trauma: 2021 Update

Head trauma (ie, head injury) is a significant public health concern and is a leading cause of morbidity and mortality in children and young adults. Neuroimaging plays an important role in the management of head and brain injury, which can be separated into acute (0-7 days), subacute (<3 months), then chronic (>3 months) phases. Over 75% of acute head trauma is classified as mild, of which over 75% have a normal Glasgow Coma Scale score of 15, therefore clinical practice guidelines universally recommend selective CT scanning in this patient population, which is often based on clinical decision rules. While CT is considered the first-line imaging modality for suspected intracranial injury, MRI is useful when there are persistent neurologic deficits that remain unexplained after CT, especially in the subacute or chronic phase. Regardless of time frame, head trauma with suspected vascular injury or suspected cerebrospinal fluid leak should also be evaluated with CT angiography or thin-section CT imaging of the skull base, respectively. The American College of Radiology Appropriateness Criteria are evidence-based guidelines for specific clinical conditions that are reviewed annually by a multidisciplinary expert panel. The guideline development and revision include an extensive analysis of current medical literature from peer reviewed journals and the application of well-established methodologies (RAND/UCLA Appropriateness Method and Grading of Recommendations Assessment, Development, and Evaluation or GRADE) to rate the appropriateness of imaging and treatment procedures for specific clinical scenarios. In those instances where evidence is lacking or equivocal, expert opinion may supplement the available evidence to recommend imaging or treatment.

Validation of the Revised Neuroimaging Radiological Interpretation System For Acute Traumatic Brain Injury in Adult and Pediatric Population

Aim Our study aimed to validate the revised neuroimaging radiological interpretation system (NIRIS), which would standardize the interpretation of noncontrast head CT of acute traumatic brain injury (TBI) patient and consolidate imaging finding into ordinal severity categories that would not only inform specific patient management actions but could also be used as a clinical decision support tool.

Methods We retrospectively studied dispositions and their outcomes of consecutive patients brought to the Sawai Man Singh Hospital Trauma Centre, Jaipur, India, by any means of transport and who underwent a noncontrast CT scan for suspected TBI between April and December 2018.

Results The revised NIRIS correctly predicted disposition and outcome in 62.9% (750/1192) of patients. After excluding patients with OMEI (other major extracranial injuries) and OMII (other major intracranial injuries), a correct prediction was observed in 88.3% (670/758) of patients. After excluding OMEI and OMII, the predictability of revised NIRIS in the adult population is 87.6% (446/509), while predictability in the pediatric population is 92.1% (224/249).

Conclusion Revised NIRIS is a good tool for predicting patient dispositions, to specific management categories, and outcomes in TBI patients after noncontrast CT head.