T1-Weighted Subtraction Images for Volumetric Assessment of Extent of Resection of Diffuse Low-Grade Gliomas
A M Ashour, M Maamoun, M Nosseir
A M Ashour, M Maamoun, M Nosseir. T1-Weighted Subtraction Images for Volumetric Assessment of Extent of Resection of Diffuse Low-Grade Gliomas. The Internet Journal of Neurosurgery. 2022 Volume 17 Number 1.
Background: The extent of resection (EOR) of Diffuse Low Grade Gliomas (DLGG) is often challenging to assess due to post-operative hyperintensity of the resection cavity and artifacts on T2-Weighted (T2W) and Fluid-Attenuated Inversion Recovery (FLAIR) sequences. In this study we propose a novel technique using OsiriX for evaluation of the postoperative EOR of DLGG.
Methods: Twelve patients with DLGG were included in the study. All lesions were hypointense on T1-Weighted (T1W) and hyperintense on T2W and FLAIR sequences. A publicly available post-processing radiographic software “OsiriX” was used to digitally subtract postoperative T2W from T1W sequences. Volumetric analysis for EOR was calculated based on the volume calculated from the subtraction images. The volumes of tumor residual and qualitative EOR calculated by this technique were compared to the volumes based on T2W images only, which were interpreted by a neuroradiologist.
Results: Based on the two imaging techniques for residual tumor volume, 1 patient was classified as Subtotal Resection (STR) according to T2W volume and Gross Total resection (GTR) according to digital subtraction T1-T2 volume. Three patients were classified as Partial Resection (PR) according to T2W volume and STR according to T1-T2 volume. Two were classified as PR according to T2W volume and STR according to T1-T2 volume (Table 1). The mean difference between the two groups was 5.6cm3[2.3-8.9]. The two groups were statistically equivalent (p = 0.05).
Conclusion: Digital subtraction of postoperative T2W from T1W sequences, combined with orientation to T2W sequence proved to be easier, reliable and reproducible in assessing the extent of resection in DLGG. This technique should be helpful in the management and follow-up of this pathology after surgical resection.
The extent of resection (EOR) of a Diffuse Low Grade Glioma (DLGG) plays a key role in determining the patient’s postoperative management. The EOR is usually subjectively assessed based on the neurosurgeon’s operative report, which seldom correlates with postoperative imaging studies. 17 In general, it is easy to distinguish high-grade glioma remnants on MRI due to contrast enhancement. However, when it comes to DLGG, distinguishing the tumor borders from the surrounding edema proves difficult. In order to overcome this limitation, we used an open-source Digital Imaging and Communications in Medicine (DICOM) viewer, OsiriX MD (Pixmeo, Geneva), to develop a specific technique able to distinguish tumor remnants from the surrounding tissue. A volume calculation method based on 3D volume rendering-assisted region-of-interest (ROI) computation has been shown to be very helpful for volume assessment. 24,33,40 The purpose of this study is to introduce a way to measure the residual volume of DLGG on post-operative MRI by utilizing the OsiriX MD DICOM viewer.
Materials and Methods
Preoperative and Postoperative Imaging
In an attempt to validate the technique, and after obtaining approval of the local institutional review board at our university hospital, the maintained database was retrospectively reviewed for patients presenting with supratentorial DLGG who underwent surgery. Patient inclusion criteria were (1) pathology confirmed DLGG and (2) available pre- and postoperative imaging. Exclusion criteria were (1) high-grade pathologies, (2) intense contrast enhancement lesions, (3) infratentorial lesions, (4) lack of adequate imaging, (5) patients who refused surgical resection. those who had previous surgery, radiosurgery or chemotherapy.
Preoperative and postoperative imaging studies of all included patients were reviewed. All volume measurements were done in cm3. Preoperative volume was calculated based on T2-weighted (T2W) imaging. For postoperative volume, manual evaluation of T2W vs. semi-manual evaluation using T1W – T2W were compared. For the manual group, postoperative T2W were used to calculate the postoperative residual tumor (VolAresidual). For the semi-manual group, postoperative T2W were subtracted digitally from postoperative T1W images to calculate the postoperative residual tumor (VolBresidual).
Synapse 3.1.1, FUJIFILM Medical Systems USA, was used to outline and calculate the preoperative volume and the postoperative residual volume in the manual group; the areas for each slice of hyperintensity in region of interest (ROI) in T2W sequence were segmented manually with orientation to FLAIR sequence of same slice to avoid hyperintensity artifacts, like resection cavity and edema (Figure 1) then multiplied by the slice thickness to calculate each slice volume. Inter-slices volume was calculated by the median of the surface area of each two successive slices multiplied by gap thickness, and then volumes of slices and inter-slices were added together to have the final volume (VolAresidual).
Semi-automated Segmentation Method
OsiriX software was used to digitally subtract postoperative T2W from T1W sequences (Figure 2) for the semi-manual group, using the included subtraction tool (Figure 3). Resample, registration, and image fusion tools were used when needed, to readjust the two sequences together if a difference in cut thickness and orientation was found (Figure 3). Upon observation of the original T2W sequences, the isointensity signal in a region of interest (ROI) in the new subtraction sequence (Figure 4) was segmented automatically using grow region method once digitally chosen, using “Grow Region – 2D/3D segmentation” (Figure 5). An algorithm was set to (threshold) interval of the isointense signal in the subtraction image while being oriented to T2W image for the same cut. The algorithm can also be adjusted manually for more accurate segmentation of the different signal intensity resembling the residual tumor. The volume of the residual tumor “VolBresidual ” was computed based on calculating the residual volume from the above mentioned method after segmentation of the isointense signal in all the cuts in the subtraction image for ROI.
EOR percentage was calculated for both groups (T2W preoperative – VolAresidual or VolBresidual) / T2W preoperative x 100), and then sub-classified into Gross-Total Resection (GTR), Subtotal Resection (STR) and Partial Resection (PR). Gross total tumor resection was defined as 100% macroscopic removal of the tumor mass, subtotal (<100%, but ≥90%) and partial (<90%) resection. A comparison was done between the manual group versus the semi-manual group.
Statistical analysis was conducted using R 3.2.2 (Vienna, Austria). 37 The volume measurement, in cm3, obtained from manual evaluation of post-operative imaging and measurement obtained with the studied method (T1-T2) were compared for equivalence using the R equivalence package. 29 No assumption of normality was made and the robust t-test of Yuen 42 was used to compute the robust Two once-sided test (rTOST) for the paired values obtained using the 2 different analysis modalities. The test size alpha was set at 0.1, the magnitude of the region of similarity, epsilon, was set at 10mL. The tested H0 (null) hypothesis was that the two computed volumes are not equivalent. Intraclass Correlation Coefficient Technique was used to assess the reproducibility/reliability of the results measures using a numeric outcome.
A total of 12 cases were accrued for validation of the aforementioned volume calculation method. Examples of postoperative ROI and segmented tumor volumes generated using subtraction method are shown in (Figure 4, 5). All studies were initially hypointense in T1W, hyperintense in T2W and hyperintense in FLAIR sequences. Postoperative T2W showed the resection cavity hyperintense, residual tumor was mild hyperintensity and vessels shown as void signals. The subtraction image showed resection cavity hypointense and residual tumor isointense. This was segmented digitally using grow region method, which was easier and more accurate. Based on subtraction imaging findings for residual tumor volume; 1 patient was classified as PR according to T2W volume and GTR according to T1-T2 volume. Three patients were classified as PR according to T2W volume and STR according to T1-T2 volume. Eight were classified as PR according to T2W volume and T1-T2 volume (Table 1).
The mean difference between the two groups was 6.9 cm3, [4.1-9.8]. The two groups were statistically equivalent (p = 0.08). Intraclass Correlation Coefficient 0.76, CI 95% (0.168- 0.931).
The importance of surgical excision on survival has led to volumetric assessment being widely used, especially in studies involving glioblastoma 10,17,20,30. In several studies authors used T2W non-contrast and T1W contrast enhanced MRIs and measured the volume by guidance of T2W images 15,18,31,38.
Patients with brain tumors may be considered for several different treatment options during the course of their disease. EOR is usually based upon the neurosurgeon’s description in the operative report, which is not quantitative and often does not correlate with postoperative imaging studies 17. Lohle et al. reported measurement by MRI that included axial T1W-weighted images before and after contrast medium injection, as well as proton density and T2W images in malignant and benign brain lesions 23. Lacroix et al. reported the volume assessment of glioblastoma (GBM) patients as defined by the area of increased signal intensity on contrast-enhanced T1W images. For non-enhancing tumors, volume was defined by the area of increased signal intensity on T2W images corresponding to the defined mass lesion in a computer software program 23. There were several studies that used the same method to calculate glial tumor assessment, but most of these studies examined contrast enhancement tumors 1,2,4,6,7,9-11,13,16,17,26,27,30,32. Hatiboglu et al. used the entire area of T2W signal abnormality and segmented on FLAIR sequences for non-enhancing tumors 10. However, it is difficult to differentiate the tumor tissue from the edematous tissue in non-enhancing tumors. Ius et al. used T1W contrast enhanced and T2W images to calculate the volume difference between preoperative and postoperative images 15. Sankar et al. used T1W contrast enhanced images to calculate the volume in oligodendroglioma cases 31. Skrap et al. and Snyder et al. used T2W magnetic resonance images (MRIs) to establish preoperative and postoperative tumor volume 34,35.
In our study, T1W sequence was used to avoid the confusion between the hyperintensity signal in T2W and FLAIR sequences in assessing the residual tumor volume. Post-processing software (OsiriX) was used to digitally subtract postoperative T2W from T1W sequences and volumetric analysis for EOR was undertaken based on the volume calculated from the subtraction image. OsiriX Medical Imaging Software is an image processing application for Mac dedicated to DICOM images produced by medical equipment (MRI, CT, PET, PET-CT, etc...) and a FDA cleared 510k class II medical device, according to US Food And Drug Regulation CFR21 part 820.
Subtraction technique was used in many applications for recurrent GBM, glomus tumor, thrombosis of vein of Galen aneurysm, evaluation of sacroiliitis and liver cirrhosis 8,14,22,25,41. The subtraction image prepared by imaging processing software showed resection cavity as hypointense and the residual tumor isointense, which can be segmented digitally using “grow region” option in Osirix. As there was a clear difference between resection cavity and residual tumor in signal intensity after substraction, this made it more accurate to use the automated segmentation tool than the manual method of marking the whole lesion. Also, vessels were excluded in the segmented part due to different pixel values of enhancement in the subtracted images. Cordova et al. used subtraction technique to evaluate GBM resection, by creating post resection tumor volumes. Subtraction images were generated by subtracting white matter–normalized, spatially co-registered pre-contrast T1W images from post-contrast T1W images, to avoid the presence of T1 hyperintense postoperative blood products and resection cavity 3.
Accurate postsurgical residual tumor segmentation are typically difficult to determine, even using the most time-consuming manual segmentation methods, due to their complex morphology and the presence of T2 -hyperintense blood products and resection cavity. The subtraction tool in Osirix was used, which digitally subtract postoperative T2W from T1W corresponding pixels to result in the subtraction image. This subtraction image was clearer for the computer software to segment precise volume and is easier than the manual method.
Manual contouring was found to decrease the inter-rater and intra-rater variability of volume generation in multiple anatomic regions, but this required a great deal of time and effort 19,36,39. Sorensen et al. reported that manual contouring of tumor took an average of approximately 20 minutes per tumor to complete volume calculation, which clearly limits its routine application 36.
Semi-automated segmentation techniques have been introduced in an attempt to combine the high-level visual processing and specialized knowledge exhibited by humans with the objectivity of computers. However, most of these techniques have not gained widespread acceptance and its usage is mainly adopted by the original developers 3,5,12,21,28,39.
Automatic or semiautomatic segmentation of radiographic images relies on differences in the grey value between a region of interest and the surrounding tissue. In most linear segmentation algorithms, starting from a specific area of interest, the neighboring pixels are evaluated and are either included or excluded based on a set threshold. It is intuitive that increasing the contrast between an area of interest and the surrounding tissue would lead to a more reliable segmentation. In the specific case of low grade gliomas, these lesions are initially hypointense on T1 and hyperintense on T2, the assumption is that any residual would have similar biological behavior and therefore similar radiographic appearance. Unfortunately, post-operatively this appearance is compromised by post-operative edema and resection cavity. To overcome this limitation, subtracting post-operative T2W from T1W sequences allows diminishing the signal engendered by post-operative edema, artifacts and filling of the operative cavity with CSF. This process increases the contrast between any residual and the surrounding tissues, therefore optimizing segmentation.
The mean difference between the two groups was 6.9 cm3 [4.1-9.8]. The two groups were statistically equivalent (p = 0.08). We believe that the difference between the volumes calculated based on T2 and the subtraction image volume and the qualitative volume too, is due to two factors. First the hyperintensity signal in T2 for edema and artifacts, which was omitted using the subtraction technique then segmenting the lesion digitally based on the pixels values generated. The second factor is the method the calculation for the T2 volume was done, estimating more residual volume as the inter-slices volume was calculated based on the median between the two successive slices. Which for sure increases the estimated tumor volume; meanwhile it was calculated digitally by the software.
Our proposed digital subtraction technique was easier and reliable for evaluating the extent of resection for DLGG and provides a novel digital quantitative method in comparison to the classic manual method. Digital subtraction of post-operative T2W from T1W sequences, combined with T2W orientation was helpful in assessing the extent of resection in DLGG and avoiding confusion from other hyper-intensity signals in T2W and FLAIR studies. This technique should be helpful in management and follow-up of this pathology after surgical resection.