1	¹Department of Radiology, Mount Sinai School of Medicine, New York ,NY ²Department of Neuropsychiatry, Universidade Federal de Pernambuco, Brazil, ³Department of Pathology, Universidade Federal de Pernambuco, Brazil, ⁴ Multimagem, Hospital Albert Sabin, Recife, Brazil ⁵ New York University Medical Center, New York, NY
2	Background Despite great progress in magnetic resonance imaging (MRI) technology, the accurate diagnosis of intracranial mass lesions remains challenging.
3	"Utilization of advanced functional MRI..." Utilization of advanced functional MRI techniques can provide additional information, not available with conventional MRI¹, in the assessment of intra-axial brain masses diffusion weighted imaging (DWI) evaluates the microstructure of brain tissue by analyzing the motion of water molecules in the tissue perfusion weighted imaging (PWI) evaluates the microvasculature within the brain using relative cerebral blood volume (rCBV) magnetic resonance spectroscopy (MRS) gives information about the metabolism of the brain tissue
4	"Among intracranial masses" Among intracranial masses, infiltrative diffuse astrocytoma represents more than 60% of all primary brain neoplasms, with an incidence of 5-7 new cases per 100,000 persons per year.² Tumor classification is made by the pathologist after a sample of the lesion is obtained by invasive biopsy or surgery. The World Health Organization² (WHO) classification remains the standard reference for guiding therapy and predicting prognosis. Astrocytomas are classified as grades I, II, III, and IV, depending on a combination of histological criteria (hypercellularity, cell and nuclear pleomorphism, mitoses, vascular proliferation, necrosis)². Higher grade tumors are more aggressive and carry worse prognosis than lower grade tumors².
5	Purpose The purpose of this study is to retrospectively determine the accuracy of a PREVIOUSLY PUBLISHED multi-parametric, algorithmic approach to the characterization of astrocytomas in adults, using conventional MRI as well as DWI, PWI, and MRS.
6	Materials & Methods We reviewed thirty-two patients from multiple centers who had intra-axial mass lesions with histological confirmation of diffusely infiltrating astrocytomas (WHO classification²). Between January 2004 and February 2007, these patients underwent conventional and advanced MR examinations prior to treatment. Ten cases were grade II, five cases were grade III, and 17 cases were grade IV. Grade II lesions were called low-grade, and grades III and IV high-grade. Patients with extra-axial lesions or HIV were excluded. Patients younger than 20 years old were also excluded in order to eliminate WHO grade I tumors, which enhance and can sometimes have increased Cho/NAA values³ and elevated PWI (85% of grade III-IV astrocytomas and 33% of grade I pilocytic astrocytomas have rCBV > 1.75 [our data, not yet published]). We used T1-SE post-contrast and short TE multivoxel MRS (TE=35ms) to identify the maximal choline/N-acetil-aspartate ratio (Cho/NAA) ratio. Regions of interest (ROIs) were chosen from any location inside non-enhancing tumors, or from the parenchyma just beyond the enhancing margin of enhancing tumors. We also evaluated PWI to determine maximal rCBV. Our results were compared with an algorithm to classify intra-axial masses, previously published by Al-Okaili, et al.⁴
7
8
9	Statistical Analysis We calculated the overall accuracy of using conventional MRI and advanced techniques, including DWI, PWI, and MRS, in accordance with the algorithm described above. Additionally, we calculated the accuracy of each question independently. Accuracy is defined as the number of correctly classified “true” cases (true positives plus true negatives) divided by the total number of cases.
10	Results *Q5 attempts to distinguish between high-grade glioma and metastasis based on Cho/NAA values in the peri-enhancement area, with a ratio of 1.0 as the threshold. This node had an accuracy of 66.7% (11/15). See following table. † These results may be affected by histopathologic undergrading of two tumors due to limited specimen size (sampling bias). The pathologist in our study classified them as grade II based on the available specimen. However, we believe that these tumors were high grade based on their high perfusion values.
11	Results: Accuracy of each question
12
13
14
15
16
17	"The integrated algorithmic MR imaging..." The integrated algorithmic MR imaging strategy described by al-Okaili was 65.6% (21/32) accurate in distinguishing diffusely infiltrating astrocytomas within our patient population (examples 1,2). The single factor that most affected accuracy appears to be Question 5, which incorrectly identified up to 27.7% (4/15) of high grade astrocytomas as metastases (example 3). Overall this question had an accuracy of 73.3% (11/15). If Question 5 is omitted, and one considers only low-grade versus high-grade neoplasms, the accuracy of the algorithm increased to 78.1% (25/32). The Al-Okaili et al.⁴ study included 40 patients with high- and low-grade neoplasms, lymphomas, tumefactive demyelinating lesions, abscesses, and encephalitis. Of those patients, only 19 had high-grade or low-grade neoplasms. Their study showed that the overall accuracy in distinguishing between high-grade and low-grade neoplasms was 90%, which was higher than our results, possibly because their series included not only astrocytomas, but also metastases and lymphomas.
18	"There are many issues affecting..." There are many issues affecting the reproducibility of histopathology diagnosis, such a intra- and interobserver subjectivity, insufficient size of tissue sample to characterize the most malignant part of the tumor (sampling bias), and frequent dedifferentiation into more malignant grades over time^14-16. We suspect that sampling error may have occurred in the histopathology classification of two lesions in our study, where inadequate tissue sample may have compromised tumor grading. On imaging, those lesions, according to the experience of the examining neuroradiologist, were more likely to be grade IV than grade II lesions (example 5). Therefore, if one assumes that the grade was underestimated and are truly high-grade, the accuracy of the algorithm would be 71.8% (23/32).
19	Conclusion In evaluating astrocytomas with algorithm described for Al-Okaili et al.⁴ the peri-enhancement Cho/NAA values, using a threshold of 1.0, were not able to consistently classify high grade lesions as infiltrative gliomas rather than metastases. Radiologists could work closer with neurosurgeons and neuropathologists not only to detect the tumor and identify the optimal biopsy site, but also to participate in the classification of brain tumors. This would likely minimize sampling errors and underestimating the biology of tumors, allowing the optimal treatment plan to be implemented. In the evaluation of intra-axial brain tumors, analyzing MR perfusion, diffusion, and MRS, along with conventional MRI, could increase the accuracy of tumor classification. Each modality provides different information that reflects the histopathologic features used in tumor classification (MRI-macroscopy, DWI-cellularity, PWI- neovascularity, Cho-cell multiplication, Lipid-Lactate-necrosis).
20	References
21