Lesion Volume Estimation from TBI–MRI

Abstract

Traumatic brain injury (TBI) is a major problem affecting millions of people around the world every year. Usually, TBI results from any direct or indirect physical impact, sudden jerks, or blunt impacts to the head, leading to damage to the brain. Current research in TBI is focused on analyzing the biological and behavioral states of patients prone to such injuries. This paper presents a technique applied on MRI images in estimation of lesion volumes in brain tissues of traumatic brain-injured laboratory rats that were subjected to controlled cortical impacts. The lesion region in the brain tissue is estimated using segmentation of the brain, diffusion, and the damage regions. After the segmentation, the area of the damaged portion is estimated across each slice of MRI and the combined volume of damage is estimated through 3D reconstruction.

1 Introduction

Traumatic brain injury (TBI) is the damage to the brain caused by external impacts such as injuries resulting from vehicle accidents, accidents in sports, sudden jerks, blunt impacts. Over 2 million people around the world are severely affected by TBI every year. A TBI can lead to death or a temporary or permanent impairment in behavioral, cognitive, and emotional activities. Current research in this field is focused on associating behavioral patterns of patients with physical injuries, analyzing tissue recovery, investigating how they compensate for memory loss, and also growing stem cells in the lesion regions [1–3]. In the current work, experimental animals, laboratory rats, were subjected to control cortical impact (CCI), where physical impacts are made on the brains, thereby damaging the brain tissues. These TBI rats are scanned periodically using magnetic resonance imaging (MRI), and the damage volumes were estimated through segmentation and reconstruction of the brain. These estimations can further help analyzing the recovery in rats and also identify other behavioral parameters associated with the damage [4].

The advent of magnetic resonance imaging (MRI) technology makes it possible to observe the direct physical condition of the patient (in vivo). It is becoming increasingly popular for its ability to detect disease without exposing patients to hazardous radiation like in X-Ray and CT imaging. Two types of mechanisms, the spin–lattice T₁ and spin–spin T₂, are used in the imaging process in generating two different contrasting tissue images [1]. The obtained images are analyzed, in separating the various regions of interest. We employed an elimination strategy, where the region of interest was separated and image metrics were obtained for area and volume of the damaged region.

The lesion volumes for estimating the damage were computed from the MRI images scanned at Bio-Imaging Research Center (BIRC), UGA. BIRC at University of Georgia has a 7T Varian System and is capable of performing various imaging methods. The facility uses image analysis tools like NUTS, jMRUI, Image J (NIH), and FSL for data and image analysis. The current paper details the strategies employed in automating the process of determining the damage volume from the MRI data of TBI rats.

The experimentation and surgical procedures followed are discussed in Sect. 2, followed by details of the segmentation process used to differentiate the regions of interest and the estimation of physical parameters. Procedures involved in the 3D reconstruction of the brain are discussed in a subsequent section.

2 Experimentation

2.1 Subjects

Male Sprague–Dawley rats (200–250 g, Harlan Laboratories) received a craniotomy operation followed by a direct injury to the cortex. All procedures involving animals strictly followed the guidelines set forth in the Guide for the Care and Use of Laboratory Animals (US Department of Health and Human Services, Pub no. 85-23, 1985) and were approved by the University of Georgia Institutional Animal Care and Use Committee.

2.2 Controlled Cortical Impact on Rats

In order to study the effect of TBI, the experimental laboratory rats were subjected to repeatable direct injury to the left fronto-parietal cortex via a controlled cortical impactor (CCI) [5, 6]. Briefly, each rat was placed under anesthesia using 5% isoflurane for induction, and its head was then quickly shaved using an electric shaver. The animal received a continuous flow of 2% isoflurane and oxygen via a fitted anesthesia mask after transferring to a stereotaxic frame (David Kopf Instruments, CA). The animal’s body temperature was kept warm using a controlled heating blanket, and the breathing rate was monitored throughout the procedure. The shaven head surface was sterilized, and a single incision was made slightly to the left of the sagittal suture. The skin and muscle tissue were retracted, and the skull was cleaned with 3% hydrogen peroxide followed by saline as previously described [7, 8]. A 5-mm craniotomy was made using a 5-mm (diameter) trephine bur on an electrical drill, at 0.5 mm from coronal suture, and 0.5 mm from sagittal suture on the left parietal bone. The bone disk was removed, and the area was cleaned with a thin piece of gel foam (PFIZER, New York, NY) saturated with saline. The dura was evaluated for any ruptures and for any bleeding. The brain was covered with the gel foam, while the CCI was quickly set up for impact. Prior to starting the surgery, the velocity (2.25 m/s) and dwell time (250 ms) were calibrated and tested. The 3-mm tip was sterilized and screwed onto the pistol of the CCI. The tip was then lowered, and the gel foam was moved to the side, to allow the tip to come in direct contact with the dura. The tip was retracted, and the gel foam was placed back on the open brain area. The retracted tip was lowered 2 mm, in order to cause a 2 mm deep injury. The gel foam was quickly moved to the side, and the CCI was fired. After the impact, the gel foam was placed over the bleeding brain, and the extent of hematoma and herniation of the brain was recorded. The speed of the pistol was calculated and recorded into each animal’s surgical sheet, a couple of minutes after the bleeding stopped; additional gel foam was used to cover the craniotomy followed by sealing with UV curing dental cement. The skin was sutured back after the hemostats were removed. The animal was given a subcutaneous injection with buprenorphine HCI and allowed to recover under a heat lamp in a new cage before it was returned to the animal room.

2.3 Brain Tissue Collection

Four weeks post-TBI, the animals were anesthetized using ketamine (50 mg/kg)/xylazine for gross anatomy analysis. The brains were carefully removed, and each extracted brain was post-fixed overnight in 4% PFA. The next day, each brain was placed in 30% sucrose and imaged under a dissecting microscope in order to capture the entire brain superficial plain. Each brain was placed on a brain matrix, and a coronal incision was made down the center of the injury. Images of the coronal view were also taken to access the extent of the gap or area voided of brain tissue.

2.4 Scanning in MRI

Four weeks after receiving a brain injury, each rat MR received an MRI scan as previously described [9, 10]. Briefly, each animal was kept under 2% isoflurane anesthesia in oxygen-enriched air via a built in nose cone and placed directly on the animal tube coil assembly. The animal’s head was securely positioned directly under the 20-mm dual-tuned surface coil, and the animal tube assembly was placed inside the 210-mm horizontal bore of the small animal 7-T MRI system (Varian Magnex). Each animal received coronal scans of the head, with each slice having a thickness of 1 mm and zero distance between each slice. The entire injured tissue was imaged and with the addition of two extra slides added to the number of slices needed to capture the entire injury, one at the rostral end and one at the caudal end. T1-weighted (T1WI) and T2-weighted (T2WI) images were collected for injury volume analysis. For T1-weighted MRIs, animals were scanned using (TR) = 0.50 s, echo time (TE) = 14 ms, a matrix size of 256 × 256 with a field of view (FOV) = 40 mm × 40 mm were chosen. The total acquisition time with two averages per phase—encode step was 4 min 17 s per rat. While, T2 images had 2.0 s, 32 ms, and 40 mm × 40 mm for TR, TE, FOV, respectively, with a 256 × 256 matrix. A total acquisition time of 4 min 20 per rat was fixed on the rats with 4 averages per phase encoding step.

2.5 Quantification of Cortical Tissue Lost from Brain Ex Vivo Images

Brain images taken of the superior view of the injured brains, along with coronal images of the injured left fronto-parietal cortex, were analyzed using Image J (NIH) to calculate the approximate total tissue loss. The images collected from the superior view of the injured brains were used to calculate the loss of tissue by manually outlining the large area of the cortex where cortical tissue was missing. The coronal view images were used to determine the total depth or thickness of the gap left by the missing tissue. The volume of cortical tissue missing was estimated by multiplying the total superior area by the depth of the injury. The results of the quantifications are illustrated in the graph below, along with sample images of the ex vivo brain superior view (A) and coronal view (B).

MRI, a noninvasive medical imaging technique, uses magnetic fields and radio waves, where a powerful magnetic field is made to oscillate at a high frequency, generating oscillations in hydrogen molecules (protons) in water. These oscillating hydrogen atoms generate radio frequency pulses which are transformed to images. Hence, tissues with higher water content hold their energy for a longer time, extending the duration of emission of radio waves. MRI images are used for identifying different anatomical structures or pathologies in tissue via weighted image contrast. After excitation, the differences in tissue content determine the contrast, or different gray levels, obtained by the independent processes of T1 (spin–lattice) and T2 (spin–spin) relaxation as tissue returns to its equilibrium state. The T1-weighted images are obtained when MR signal is measured after magnetization is allowed to recover through repetition time (TR). Cerebral cortex assessment, general morphological information identification, and post-contrast imaging can be done through this image T2-weighting. On the other hand, the image magnetization in T2-weighted imaging is allowed to decay prior to MR signal measurement for a varying echo time (TE). This method is useful for detecting edema, white matter lesions, hypoxia, inflammation and assessing hypoxic ischemic changes in brain injury. Additionally, chronic consequences of brain injury such as diffuse atrophy and gliosis have been identified using T2 MRIs [11]. T2 MRIs tend to be the most common MRIs performed to assess brain injuries because of the broad spectrum of different tissue injuries that can be observed using T2 MRIs. While the T2 MRI images allow for an overall qualitative observation of the injury, the need for a method to accurately and reliably quantify the volume of injured tissue continues to be unmet. In order to address this issue, we developed a method to efficiently quantify hyper-intensity in T2 MRIs. Figure 1 shows the scanned portions of the brain, and Fig. 2 shows coronal brain images of a T2 MRI scan obtained from one of the rats. MATLAB was used for converting the image formats and in separating layers of MRI images. The regions of interest are separated applying the designed algorithm. As shown in Fig. 2, the diffusion (white) and damaged (blue) region areas were computed in every image and all the images were combined for estimating the volume of damage. The images of different layers of MRI images are presented in Fig. 2 (Fig. 3).

Fig. 1.

Coronal view images used to determine the total depth or thickness of the gap left by the missing tissue

Fig. 2.

Region of interest in MRI images of TBI rats

Fig. 3.

MRI scanning of TBI rats

3 Implementation and Results

This section describes the procedures followed in separating the region of interest from the rest of the image. First, the brain region is separated from the skull and muscle portion surrounding it. It must be noted that a healthy brain is more or less evenly hydrated and represents a narrow band of intensities in the image. The skull and other muscle portions, on the other hand, contain lesser water content and hence display lower intensities. The algorithms implemented were taken from [12].

Also, the diffusion portion which is largely composed of water shows very high intensity in the images, and hence, it is easy separating diffusion portion from the healthy portion of the brain. In order to compute the damage area, the diffusion inside the brain region is computed, discussed later on in this section.

The separated regions are summed up for the number of pixels, and their corresponding areas are computed. The overall algorithm flowchart is presented in Fig. 4.

Fig. 4.

Algorithm overview

3.1 Segmenting Brain Region

Each of the MRI images separated is subject to adaptive threshold technique. The threshold value obtained usually ensures that the brain region and the skull region are separated, as the number of pixels in the brain region is higher.

The separated brain region, however, contains unwanted regions falling within the threshold limits. Hence, the image is labeled using connected component labeling (CCL). This allows the separation of the regions which are large in size. The detailed segmentation of the image is presented in the flowchart in Fig. 5.

Fig. 5.

Separation of brain and skull in MRIs

3.2 Segmenting Diffusion

The images within the brain region are further segmented and processed for separating the diffusion region. The diffusion portion is usually at a higher intensity, and its threshold generally falls above 200. Hence, the regions with intensities higher than the threshold are separated, and the largest connected portions are selected through CCL. The segmentation procedure is shown in Fig. 6.

Fig. 6.

Segmentation of diffusion region

3.3 Injury Identification

The separated brain and diffusion segments of the image are raster scanned in parallel. In every row, the number of high pixels (1s) in the brain segment should always be greater than the diffusion segment high pixels. Otherwise, the pixels in that row are made zero. This gives the approximate diffusion region falling inside the brain region, which is nothing but the damaged portion. The flowchart for the algorithm is presented in Fig. 7.

Fig. 7.

Separation of injury

3.4 Area and Volume Computation

Each of the images in the MRI scan image constitutes 256 × 256 pixels corresponding to 4 mm × 4 mm of tissue. The area constituted by n pixels is given by

$$\mathit{\text{Area}}=n\times \frac{1600}{250\times 256}=n\times 0.02441{\text{ mm}}^2$$ (1)

wheren is the number of pixels of interest. The process is repeated on all the layers of MRI images. All the areas are summed up and multiplied with the layer’s thickness for computing approximate volume. The segmented data obtained after applying the algorithm is reconstructed on to a 2D image as shown in Fig. 8. The volume estimation and the 3D reconstructed model for a sample MRI are presented in Fig. 9.

Fig. 8.

2D reconstruction

Fig. 9.

2D, 3D reconstruction and area and volume estimation form a sample MRI of a TBI rat

$$\mathit{\text{Volume}}=\sum _{i=1}^L{n}_i\times 0.02441\times S{\text{ mm}}^3$$ (2)

where

• i Number of layer/MRI image

• L Total number of MRI images

• n_i Number of pixels of interest in ith layer

• S Thickness of the layer

4 Conclusions

We presented an algorithm for segmenting brain diffusion and injury in T2 images. We prototyped it for preliminary analysis of rat TBI–MRI images and estimated the area and volume of damage. Further, 2D and 3D reconstructions of the brain were done for visualization. Since the method presented here is limited to the intensity differences in the T2 MRIs which indicate a broad spectrum of injury at the edge of the cortex, additional noise created by the craniotomy and brain herniation needs to be accounted for in order to eliminate non-cortical areas. However, this quantification method can be applied to scans of more central areas of the brain and/or to other types of scans designed to identify more specific changes in tissue.