Automated Segmentation and Severity Analysis of Subdural Hematoma for Patients with Traumatic Brain Injuries

Abstract

Detection and severity assessment of subdural hematoma is a major step in the evaluation of traumatic brain injuries. This is a retrospective study of 110 computed tomography (CT) scans from patients admitted to the Michigan Medicine Neurological Intensive Care Unit or Emergency Department. A machine learning pipeline was developed to segment and assess the severity of subdural hematoma. First, the probability of each point belonging to the hematoma region was determined using a combination of hand-crafted and deep features. This probability provided the initial state of the segmentation. Next, a 3D post-processing model was applied to evolve the initial state and delineate the hematoma. The recall, precision, and Dice similarity coefficient of the proposed segmentation method were 78.61%, 76.12%, and 75.35%, respectively, for the entire population. The Dice similarity coefficient was 79.97% for clinically significant hematomas, which compared favorably to an inter-rater Dice similarity coefficient. In volume-based severity analysis, the proposed model yielded an F1, recall, and specificity of 98.22%, 98.81%, and 92.31%, respectively, in detecting moderate and severe subdural hematomas based on hematoma volume. These results show that the combination of classical image processing and deep learning can outperform deep learning only methods to achieve greater average performance and robustness. Such a system can aid critical care physicians in reducing time to intervention and thereby improve long-term patient outcomes.

Automated Detection and Screening of Traumatic Brain Injury (TBI) Using Computed Tomography Images: A Comprehensive Review and Future Perspectives

“…Contrast limited adaptive histogram equalisation (CLAHE) is also utilised as a technique for enhancing image quality [ 51 ]. Moreover, clustering techniques, such as K-means [ 52 ], Fuzzy c-Means (FCM) [ 53 , 54 , 55 , 56 ], and level-set methods, such as the distance regularised level set evolution (DRLSE) [ 57 , 58 , 59 ], are used for region of interest (ROI) extraction.…”

“…Texture feature extraction is widely used as a feature extraction technique, and the inclusion of shape features [ 67 ] and statistical features have proven useful for improved diagnosis of TBI [ 44 , 51 , 57 , 68 , 69 ]. Raghavendra et al [ 70 ] extracted a set of nonlinear features based on entropy to detect the presence of intracranial hematoma in CT images and obtained an accuracy of 97.37%.…”

“…The hematoma segmentation in CT imagery can be realized using rule-based models [ 23 , 41 , 76 ] or machine learning models [ 18 , 27 , 57 , 58 , 69 , 77 ]. Chan [ 23 ] developed a knowledge-based classification system from symmetry analysis to detect acute hematoma.…”

“…The classification task involves extracting a set of relevant features initially to classify each pixel/voxel as hematoma, whereupon suitable post-processing techniques are then applied to obtain fine segmentation. The post-processing techniques include thresholding, morphological operations, smoothing, and active contours [ 18 , 27 , 57 , 58 , 69 , 77 , 78 ]. Farzaneh et al [ 58 ] used handcrafted statistical, textural, and geometrical features to classify each super pixel as normal or SDH.…”

“…Farzaneh et al [ 58 ] used handcrafted statistical, textural, and geometrical features to classify each super pixel as normal or SDH. The same research group [ 57 ] has incorporated deep features in their recent study to improve segmentation performance. Scherer et al [ 68 ] extracted statistical and textural features to perform voxel-wise hematoma classification.…”