U-Net architecture shows higher sensitivity and precision for subdural hematoma detection in meta-analysisMachine learning improves detection of brain bleeding in CT scans

Neurosurgical review Published August 7, 2026 Study authors: Dumasia Suraj R, Ahmed Warda, Edavettal Michelle, Castellanos Mackenzie, Gandhi Om H, Lanzino Giusep… PubMed ↗ DOI ↗ Editorial oversight: Dr. Lars van Dijk, PhD · Surgical, Procedural & Diagnostic

AI-generated summary of the cited source, checked by automated accuracy review. How we work

Key Takeaway

Note that U-Net architectures may offer superior sensitivity and precision for subdural hematoma detection on CT scans.

This meta-analysis evaluated the performance of various machine learning models, including Convolutional Neural Networks (CNN), U-Net architectures, and hybrid deep learning models, for detecting subdural hematoma on non-contrast CT scans. The analysis included 67,266 scans across 30 testing datasets to compare diagnostic metrics like sensitivity, specificity, accuracy, and diagnostic odds ratio (DOR).

The synthesis indicates that U-Net architectures demonstrated significantly higher sensitivity (0.916; p=0.04) and precision (0.983; p=0.001) compared to other models. While high specificity and accuracy were consistently observed across all deep learning techniques, the influence of internal testing on specificity was borderline significant (p=0.05). Additionally, U-Net architecture was a borderline significant predictor for diagnostic odds ratio (p=0.049).

A primary limitation noted by the authors is that the findings are based on only 4 pooled U-Net datasets compared to 22 pooled CNN architectures. Consequently, definitive conclusions regarding the superiority of U-Net over other designs are limited. These results suggest potential advantages for U-Net models in clinical settings, but more well-powered studies are required to confirm these performance differences.

When a patient suffers a head injury, doctors must quickly identify internal bleeding called a subdural hematoma. This scan is vital for making fast decisions about care. A large review of over 67,000 CT scans looked at how different machine learning models perform in spotting these bleeds.

The study compared several types of computer models. It found that a specific design called U-Net showed significantly higher sensitivity and precision than other methods. Sensitivity means the ability to correctly identify a bleed when it is actually there, while precision measures how often the model is correct when it flags a problem.

While these results are promising for improving diagnostic accuracy, the researchers noted some limitations. The study only compared four U-Net datasets against 22 other types of models. Because of this smaller sample size for the U-Net group, more large-scale studies are needed to confirm if it is truly superior to all other designs.

What this means for you:

U-Net machine learning models show higher sensitivity and precision in detecting brain bleeds on CT scans.

Common questions

What is a subdural hematoma?

A subdural hematoma is a collection of blood between the skull and the surface of the brain. It often happens after a head injury. Detecting it quickly on a CT scan is critical for medical treatment.

How accurate are these machine learning models?

The study found that all deep learning techniques showed high accuracy and high diagnostic odds ratio values. Specifically, the U-Net architecture showed significantly higher sensitivity (0.916) and precision (0.983) compared to other models.

Is one model better than all others?

The study suggests that U-Net models may be superior in sensitivity and precision, but researchers caution that this is not a definitive conclusion. This is because they only had four U-Net datasets to compare against 22 other types of models.

Study Details

Study typeMeta analysis

EvidenceLevel 1

PublishedAug 2026

PMID42557478

View Original Abstract ↓

Manual evaluation of non-contrast CT scans (NCTS) for detecting subdural hematoma (SDH) is time consuming, potentially inaccurate, and subjective to the expert analyzing them. In recent years, two deep learning (DL) algorithms have been popularly studied in this respect, namely convolutional neural networks (CNN) and U-Net architectures, the latter being a specialized type of CNN. We performed the first meta-analysis comparing various DL models for SDH detection. MEDLINE, Cochrane, Scopus, and Embase databases were searched from inception through December 2025. Studies evaluating ML model performance on an independent test dataset were included. The main outcome measures were sensitivity, specificity, diagnostic odds ratio (DOR), accuracy, and precision of CNN, U-Net, and hybrid DL models. Univariate meta-regression analyses were performed. 30 testing datasets incorporating 67,266 NCTS were included. U-Net demonstrated significantly higher sensitivity (0.916;p = 0.04) and precision (0.983;p = 0.001) while high specificity, DOR, and accuracy values were consistently observed across all DL techniques. Internal testing (p = 0.05) was a borderline significant predictor of high specificity while recent publication year (p < 0.001), U-Net architecture (p = 0.035), and 3D models (p = 0.022) emerged as significant moderators of high precision. The U-Net architecture was also a borderline significant predictor of high DOR (p = 0.049). While this single arm meta-analysis depicts potential superiority of U-Net models with respect to sensitivity and precision, these findings are based off only 4 pooled U-Net datasets in comparison to the 22 pooled for CNN architectures. Future well-powered studies evaluating the U-Net model are necessary to ensure a fair comparison of U-Net architectures to other DL designs before reaching to any definitive conclusions in this respect.

U-Net architecture shows higher sensitivity and precision for subdural hematoma detection in meta-analysisMachine learning improves detection of brain bleeding in CT scans

Common questions

Study Details

Machine Learning Model Performance for Predicting Contrast Induced Nephropathy in Coronary Interventions

Machine learning models help predict kidney damage during heart procedures

Clinical research that matters. Delivered to your inbox.

U-Net architecture shows higher sensitivity and precision for subdural hematoma detection in meta-analysisMachine learning improves detection of brain bleeding in CT scans

Common questions

Study Details

Machine Learning Model Performance for Predicting Contrast Induced Nephropathy in Coronary Interventions

Machine learning models help predict kidney damage during heart procedures

Clinical research that matters. Delivered to your inbox.

Related in Radiology & Imaging

From Other Specialties