OBJECTIVE To assess the clinical feasibility and the accuracy of two pragmatic methods in comparison with a conventional computer based method of measurement of masses from CT.
METHODS Nineteen CT scans of 11 patients with severe head injury, showing 34 traumatic lesions, were examined. The volume of every lesion was digitally measured, then a panel of three examiners independently repeated the measurement using the ellipsoid and the Cavalieri method in random order.
RESULTS All the lesions were identified by all the readers and the mean volume measured by each examiner differed by less than 1.5 ml. The average reading time for each scan was 4 minutes for the ellipsoid and 7 minutes for the Cavalieri method. The average volume of the lesions was 34.2 (SD 35) ml with the digital system, and 38.4 (SD 41) ml and 34.8 (SD 36) ml for the ellipsoid and the Cavalieri readings respectively. The average difference between the applied technique and the digital system was 0.57 (SD 9.99) ml for the Cavalieri direct estimator and 0.20 (SD 15.48) ml for the ellipsoid method. The 95% confidence interval for this difference fell between –2.75 and 3.89 ml for the Cavalieri, and between –4.94 and 5.35 ml for the ellipsoid method. There were 19 lesions >25 ml; the ellipsoid method identified 16 of them, whereas 17 were classified with the Cavalieri method. When considering individual lesions rather than the average volume, discrepancies were detected with both methods. The ellipsoid method was less precise, especially when extracerebral lesions were measured.
CONCLUSIONS Both pragmatic methods are inferior to computer based reading, which is the choice when accurate volume estimation is necessary. However, if a digital volumetric determination of the lesions using a CT computer is not possible, the two pragmatic methods offer an alternative.
- head injury
- computed tomography
- mass lesions
Statistics from Altmetric.com
Due to the physical properties of the dura and the skull, every acute increase of volume in the intracranial space may lead to intracranial hypertension.1 This is especially relevant after traumatic brain injury, when the development of post-traumatic intracranial masses is the most dangerous complication.2 3
When a cerebral contusion or an extra-axial collection increases, ample variations in volume are easily detected without meticulous measurement. When subtle changes develop over time, on the other hand, the precision provided by more scientific measurement techniques offers a more valuable basis for clinical decision making.
The calculation of a given area, and the reconstruction of a delimited volume, are quite feasible with modern CT apparatus, especially when a dedicated workstation is part of the machinery. Unfortunately, volume determinations are seldom performed during routine CT readings, although the digital measurement of the lesions is the most accurate system available today. Finally, digital measurement proves impossible or cumbersome when multicentre trials are conducted, and central reading of CT is planned.
There is, therefore, a need for pragmatic methods of volume measurement on the CT printouts, provided that those methods are accurate and feasible.
The ellipsoid method (EM) was developed to calculate the volume of arterovenous malformations.6 It is based on the concept that the volume of an ellipsoid is about one half the volume of the parallelepiped into which it is placed. By measuring three diameters of a given lesion in the arterial phase of an angiogram, a parallelepiped is reconstructed, and its volume, divided in half, is close to the real volume of the malformation. By extending this geometrical concept from angiography to CT scan, calculation of space occupying lesions becomes possible (fig 1 A).
More recently, the Cavalieri direct estimator (CDE) has been introduced.7 It breaks down the lesion on CT into a corresponding number of points, through special grids; the volume of a lesion appearing on the scan is the product of the sum of the points that fall on the lesion, the area associated with each point, and the distance between scan slices (fig 1 B). The grid can be constructed by photocopying a template provided in the original article7or by preparing a uniformly spaced point grid (by computer or by hand) to be copied onto an overhead transparency.
As volume measurement is clinically important, but rarely performed in the neuroradiological practice, a posteriori methods, feasible even outside the neuroradiology department, may be useful or necessary. We have therefore assessed the performances and the feasibility of two pragmatic methods (the EM and the CDE) in measuring the volume of mass lesions in cases of patients with severe head injury.
Materials and methods
Eleven patients with severe head injury admitted to the neuroscience intensive care unit of a university hospital in 1997 were studied. They were eight men and three women, mean age 53 (SD 19) years, with a median Glasgow coma score after resuscitation of 6. Each patient underwent CT at admission, and repeated controls thereafter. Among the scans of these patients, 19 containing mass lesions were selected. A lesion was defined as a hyperdense or mixed (hyperdense and hypodense) area. Each scan was digitally processed (CT Scan Pace, General Electric, Maryland, USA) to determine the volume of every lesion. Such measurements were taken by a neuroradiologist, who traced every lesion on the screen and calculated the area and volume of interest. That was accepted as the reference gold standard. The results of these measurements were not disclosed to the panel of examiners.
Three members of the intensive care unit staff were trained in the identification and measurement of areas and volumes on the CT slices by using the two pragmatic methods. As part of the training, the examiners performed collective readings of several scans of patients not participating in the study, and repeated measurements of lesions using the CDE and the EM under the guidance of a neuroradiologist. Grids for the CDE were prepared and provided to the examiners. Relevant references in the literature were reviewed. After this training period, which required about 2 weeks, the 19 CT scans and appropriate examination forms were submitted to each panellist separately. The CDE and EM were used in a random order.
Both the volumes measured in each scan and the reading time for each method were recorded in the forms. The data forms were then collected, summarised, and compared with the gold standard.
To assess the agreement between the two pragmatic methods and the gold standard8 the statistical method devised by Bland and Altman9 was used. A two way analysis of variance (ANOVA) was used for measuring the interobserver variability. The Pearson χ2 test with Yates continuity correction was used for comparing percentages.
The 19 CT scans studied comprised 23 contusions/lacerations, 10 subdural haematomas, and four extradural haematomas. The range of the lesions directly computed at the CT scan varied from 3 ml to 124 ml; the sum of all lesions produced a total volume of 1260.65 ml, with an average volume for every lesion of 34.23 (SD 35.88) ml.
READER AGREEMENT AND READINGS
The readings performed by the three examiners led to consistent measurements. The results of their calculations were very close; the mean volume measured by each examiner differed by <1.5 ml. The ANOVA analysis confirmed this finding, excluding any significant difference among the readers in the use of both methods. All 37 lesions were identified by all readers, providing 222 pragmatic measurements to be compared with the gold standard.
The measurements calculated by the three readers with each method for every lesion were averaged; this average represented the volume calculated with the method. The range of the lesions measured with the EM varied from 0.9 to 147.4 ml; the sum of all lesions produced a total volume of 1273.87 ml, with an average volume for every lesion of 34.43 (SD 40.98) ml. The range of the lesions measured with the CDE varied from 4.8 ml to 150 ml; the sum of all lesions produced a total volume of 1287.42 ml, with an average volume for every lesion of 34.8 (SD 36.6) ml.
AGREEMENT OF THE TWO METHODS WITH THE GOLD STANDARD
The agreement between the CDE and EM and the computed reading was acceptable on average, but questionable in some cases.
The average difference between the applied technique and the reference value (termed bias, a measure of systematic error with the statistical method of Bland and Altman9) was 0.57 (SD 9.99) ml for the CDE and 0.20 (SD 15.48) ml for the EM. The 95% confidence interval (95% CI) for this bias fell between –2.75 and 3.89 ml for the CDE, and between –4.94 and 5.35 ml for the EM. The maximum difference between the pragmatic readings and the reference value was in the order of 25 ml with the CDE and reached 35 ml with the EM.
A graphical description of the agreement between each pragmatic method and the gold standard is illustrated in figs 2 and 3. In these plots the distance between the line of equality, lying on the zero difference between the pragmatic method and the reference reading, expresses the discrepancy between each measurement and the true value. There are discrepancies among the two pragmatic methods and the digital recording in individual cases. With the CDE these discrepancies are few and more evident at large volumes; with the EM the discrepancies are more frequent and present at lower volumes.
The limits of agreement, an additional mathematical descriptor of the agreement, were between –19.41 and 20.55 ml with the CDE and varied between –30.76 and 31.16 ml with the EM.
MASSES GREATER THAN 25 ML
Since a volume greater than 25 ml is considered a threshold for surgical indication, the capability of appropriately identifying this cut off value may indicate the clinical reliability of pragmatic methods. Nineteen lesions >25 ml were found from the CT analysis. Seventeen lesions >25 ml were identified with the CDE and 16 were detected with the EM. There is not a statistical significant difference between these two findings.
EXTRACEREBRAL AND INTRACEREBRAL LESIONS
Volume may become more difficult to measure in lesions of a particular shape. Extracerebral lesions, such as subdural haematomas for example, may be more difficult to measure than a rounded contusion in the brain parenchyma. The performances of the two pragmatic methods in determining the volume of intracerebral or extracerebral lesions were, in fact, different. The CDE was more accurate than the EM, especially in cases of extracerebral masses. The main data are reported in the table.
READING TIME AND LEARNING EFFECT
The average reading time to complete the measurement of a CT scan was 7 minutes 44 seconds (SD 42 seconds) for the CDE method and 4 minutes 27 seconds (SD 38 seconds) for the EM (p<0.001). As a reference, the average reading time using the digital system was 2 minutes 54 seconds (SD 26 seconds).
In both methods the reading time for each CT was higher at the beginning of this research and decreased as the study proceeded. Such decrease was probably due to a learning effect, as the accuracy of the measurement did not change over time. The examiners became more experienced and read the CT faster, but not at the price of reducing the precision of their analysis. The reduction in reading time was 37% with the CDE method, and 48% with the EM.
The volume of intracranial lesions, and its change over time, is of obvious interest in the diagnosis and management of patients with head injury. Specific features of space occupying lesions, including their volume, are associated with the outcome of patients with traumatic brain injury.6
Based on data from the traumatic coma data bank,10Marshall et al 5 showed that the existence of specific findings from CT, including the presence of masses exceeding 25 ml, was associated with increasing degrees of mortality and morbidity. This scale is now widely used both in clinical practice and in the scientific literature on head injury. Interestingly, this scale requires the measurement of the volume of intracranial lesions, as the threshold of 25 ml is part of the grading.
There are clinical problems, such as the evolution of a cortical contusion, that require volume measurement. The overall mass effect, which is considered for clinical management, depends on the size of a given lesion and on the associated swelling and oedema as well, but the size of the lesion itself is crucial.
For these reasons an accurate estimate or, possibly, the measurement of the volume of cerebral masses has become an integral part of the treatment and of the classification of severe head injury.
Unfortunately, the determination of the volume of masses is not performed as an element of the routine reading of CT scans. In a recent multicentre study on head injury13 involving 18 neurotraumatological centres, all 18 centres claimed to have used the Marshall scale, but the volume of lesions was seldom measured in one centre, and never measured in the other 17 (personal unpublished data).
The data collected in this study show that both pragmatic methods are on average acceptably precise when performed by readers who underwent a short training period. Accuracy was achieved by all readers, as no difference was found among them in their calculation of the volume. The systematic error of the pragmatic methods was <1 ml, and the mean volume of all lesions measured with the EM and CDE was very close to the volume measured digitally. When considering individual lesions rather than the average volume, however, discrepancies were detected with both methods and some discrepancies were relevant especially with the EM.
The CDE is more accurate than the EM. That is proved by the Bland and Altman analysis, as suggested by the comparison of figs 2 and 3. The bias is lower with the EM because this method determines both overestimates and underestimates, leading to a narrow final mean value. By comparing statistical measures of dispersion, such as SD, or specific parameters of the Bland and Altman analysis, such as the limits of agreement, it seems that the CDE provides measurements closer to the reference value.
Similarly, this method performed better in cases of extracranial lesions.
The EM was developed to assess the volume of vascular malformations and it is not surprising that it does not excel in the head injury field. The EM shows its weakness mainly when extracerebral collections are concerned. In fact, the shape of these lesions is less suitable for calculation by the ellipsoid theory. By contrast, the CDE method may accurately take into account even the irregular shape of thin lesions.
Such greater accuracy of the CDE technique is counterbalanced by the need for a longer execution time; this system takes twice as long to perform as the EM. The necessary reading time, however, is initially 8 minutes but, after a brief training, it decreases to an average of 5 minutes 11 seconds for each CT. With this reduction due to practice, the reading time of each scan with the CDE remains fully acceptable in the clinical setting.
Both pragmatic methods are inferior to computer based reading, which is the choice when accurate volume estimation is necessary. However, if a digital volumetric determination of the lesions using a CT computer is not possible, the two pragmatic methods offer an alternative.
They may be mastered after short training, do not require expensive equipment, and provide a different level of accuracy. The CDE performs better than the EM but takes longer. It is recommended whenever accuracy of measurement is crucial for possible therapeutic decisions.
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.