BACKGROUND AND OBJECTIVES: Middle meningeal artery embolization (MMAE) is effective for chronic subdural hematoma (SDH), but failure rates vary and risk prediction tools are lacking. This study developed and validated a machine learning (ML) model to predict MMAE treatment failure for SDH. METHODS: We conducted an international, multi-institutional retrospective cohort study of patients who underwent MMAE for SDH between 2019 and 2025. The cohort was split into training and test sets, stratified by treatment failure, with all hematomas from each patient assigned to a single cohort to prevent data leakage. Multiple ML algorithms were evaluated using 5-fold cross-validation, with gradient boosting selected as the final model. The primary outcome was treatment failure, defined as rescue surgical evacuation or persistent/recurrent SDH requiring retreatment. Model performance was assessed using area under the curve-receiver operating characteristic. Patients were stratified into low-risk (<5%), medium-risk (5%-15%), and high-risk (>15%) groups using a priori clinically defined thresholds. RESULTS: Among 692 treated hematomas in the test cohort, 64 (9.2%) experienced treatment failure. The gradient boosting model achieved an area under the curve-receiver operating characteristic of 0.856 (95% CI, 0.803-0.902) before calibration and 0.832 (95% CI, 0.766-0.889) after isotonic calibration, with excellent calibration (Brier score, 0.053). The most important predictors were age-platelet interaction, platelet-hemoglobin ratio, platelet count, SDH thickness, and age. Observed treatment failure was 1.6% (95% CI, 0.4%-5.7%) in the low-risk group (n = 124; 17.9%), 6.0% (95% CI, 4.3%-8.4%) in the medium-risk group (n = 516; 74.6%), and 59.6% (95% CI, 46.1%-71.8%) in the high-risk group (n = 52; 7.5%) ( P < .001 for trend), a 37.2-fold difference between high- and low-risk groups. CONCLUSION: This ML model stratified MMAE patients into risk groups with a 37.2-fold difference in treatment failure. High-risk patients had 59.6% failure, whereas low-risk patients had 1.6% failure. Risk stratification may improve patient selection and monitoring.
BACKGROUND AND OBJECTIVES: Many spinal imaging reports are written above recommended health literacy levels, limiting patient comprehension at a time when rapid electronic access to radiology results is increasing. We aim to evaluate whether supplemental natural language processing (NLP)–modified radiology reports—within the context of usual patient portal access to the original report—are associated with improved patient-reported perceived understandability compared with standard reports among patients undergoing spine imaging. METHODS: This is a prospective randomized controlled study with limited participant blinding, conducted with 2 spine surgeons affiliated with a single academic medical center in New York City. Participants (18 years or older) undergoing spine imaging at initial surgical consultation were enrolled and were randomized to receive either their original radiology report or a supplemental NLP–modified version. Radiology reports were rewritten using a standardized natural language processing prompt (ChatGPT 4.0) and reviewed for accuracy by study investigators. The primary outcome was survey-based patient-reported perceived understandability. Secondary outcomes included objective readability scores (Flesch-Kincaid Grade Level, Flesch Reading Ease, Simple Measure of Gobbledygook, Gunning-Fog Index). RESULTS: Fifty-two adults were enrolled; 42 (81%) completed the postintervention survey. Among 52 randomized participants (26 women [50%]), NLP–modified reports demonstrated significantly improved objective readability across all indices compared with original reports (Flesch-Kincaid Grade Level, 5.9 ± 1.1 vs 11.0 ± 1.7; Reading Ease, 71.0 ± 10.3 vs 28.2 ± 12.5; Simple Measure of Gobbledygook, 20.1 ± 12.7 vs 23.5 ± 15.7; Gunning-Fog Index, 8.2 ± 1.1 vs 14.3 ± 1.8; all P < .01). Of 42 survey respondents, those receiving supplemental NLP-modified reports reported greater ease of understanding the report language (4.2 ± 1.2 vs 2.6 ± 1.5, mean-difference 95% CI: 0.75-2.53, P < .01, Bonferroni-corrected- P = .02), and less difficulty with medical terms (2.0 ± 1.5 vs 3.6 ± 1.4, mean-difference 95% CI: −2.51 to −0.60, P < .01, Bonferroni-corrected- P = .04). CONCLUSION: Supplemental NLP-modified radiology reports were associated with improved objective readability and higher patient-reported perceived ease of understanding in a spine surgery initial consultation setting. These findings support further investigation of NLP-generated supplemental explanations within supervised clinical workflows.
BACKGROUND AND OBJECTIVES: Spontaneous intracerebral hemorrhage (ICH) has the highest short-term mortality among stroke subtypes, and progress in treatment has been hindered by the limited accuracy of current prognostic tools. We investigated whether white matter functional disconnection severity, estimated using a normative lesion network mapping framework integrating direct electrical stimulation and resting-state functional MRI data from previously published independent Data sets, improves prediction of 6-month functional outcome measured with the modified Rankin Scale (mRS). METHODS: We analyzed 97 patients with spontaneous ICH. Ordinal logistic regression models were used to evaluate the association between domain-specific disconnection severity and 6-month mRS, and to compare traditional functional outcome in patients with primary intracerebral hemorrhage and ICH scores with connectivity-enriched models incorporating disconnection measures. RESULTS: White matter functional disconnectivity was significantly associated with 6-month functional outcome (mRS; P < .001; Nagelkerke's R 2 = 0.507). Integrating disconnection severity into the functional outcome in patients with primary intracerebral hemorrhage and ICH scores improved predictive performance in both cases (Nagelkerke's R 2 = 0.615 vs 0.389 and 0.651 vs 0.434, respectively), with the connectivity-enriched ICH score achieving the strongest overall fit across all models tested. Exploratory analyses showed associations between domain-specific disconnection severity, residual functional deficits, and early mortality. CONCLUSION: Assessment of white matter functional disconnection severity seems to be a promising approach for predicting long-term disability after ICH, although prospective validation in larger multicenter cohorts is required before clinical implementation. The findings suggest that different brain networks exhibit variable vulnerability to injury and differential potential for functional recovery.
BACKGROUND AND OBJECTIVES: Radiation-induced meningiomas (RIMs) are the most common secondary intracranial tumors. Compared with sporadic meningiomas (SMs), RIMs exhibit more aggressive clinical behavior. Surgical resection has traditionally been the primary treatment modality for these tumors; however, previous studies have demonstrated stereotactic radiosurgery (SRS) is an effective treatment for RIMs. The aims of this study were to further characterize these rare tumors and to directly compare the outcomes after SRS vs surgical resection as the primary treatment modality for RIMs. METHODS: A single-institution retrospective analysis of surgically and SRS-treated RIMs was conducted. Patient demographic data, World Health Organization grade, tumor volume, multiplicity, complications, and extent of resection were collected. The primary outcome was local control. RESULTS: Forty-seven patients with one or multiple RIMs comprising 73 tumors were identified and compared with a cohort of 75 patients with SMs (n = 86 tumors). Patients with RIMs more frequently harbored multiple tumors (44.4% vs 16.0%; P = .002). RIMs treated surgically exhibited a significantly worse local control compared with surgically treated SMs (log-rank P = .043). Among RIMs, primary treatment with SRS was associated with a significantly better local control compared with surgery as the primary treatment modality ( P = .011). SRS demonstrated superior 5- and 7-year local control rates even after controlling for size and restricting the analysis to gross totally resected tumors. CONCLUSION: SRS was associated with better local control compared with surgical resection for the primary treatment of RIMs. Frequent multifocal presentation and high rates of treatment failure support the notion that RIMs represent a distinct biological entity. Given the long latency and ongoing risk of tumor development after cranial irradiation, these findings underscore the importance of long-term surveillance in this patient population.
