Automatic multi-modal MR tissue classification for the assessment of response to bevacizumab in patients with glioblastoma.

BACKGROUND: Current methods for evaluation of treatment response in glioblastoma are inaccurate, limited and time-consuming. This study aimed to develop a multi-modal MRI automatic classification method to improve accuracy and efficiency of treatment response assessment in patients with recurrent glioblastoma (GB). MATERIALS AND METHODS: A modification of the k-Nearest-Neighbors (kNN) classification method was developed and applied to 59 longitudinal MR data sets of 13 patients with recurrent GB undergoing bevacizumab (anti-angiogenic) therapy. Changes in the enhancing tumor volume were assessed using the proposed method and compared with Macdonald's criteria and with manual volumetric measurements. The edema-like area was further subclassified into peri- and non-peri-tumoral edema, using both the kNN method and an unsupervised method, to monitor longitudinal changes. RESULTS: Automatic classification using the modified kNN method was applicable in all scans, even when the tumors were infiltrative with unclear borders. The enhancing tumor volume obtained using the automatic method was highly correlated with manual measurements (N=33, r=0.96, p<0.0001), while standard radiographic assessment based on Macdonald's criteria matched manual delineation and automatic results in only 68% of cases. A graded pattern of tumor infiltration within the edema-like area was revealed by both automatic methods, showing high agreement. All classification results were confirmed by a senior neuro-radiologist and validated using MR spectroscopy. CONCLUSION: This study emphasizes the important role of automatic tools based on a multi-modal view of the tissue in monitoring therapy response in patients with high grade gliomas specifically under anti-angiogenic therapy.

Hemodynamic response imaging: a potential tool for the assessment of angiogenesis in brain tumors.

Blood oxygenation level dependence (BOLD) imaging under either hypercapnia or hyperoxia has been used to study neuronal activation and for assessment of various brain pathologies. We evaluated the benefit of a combined protocol of BOLD imaging during both hyperoxic and hypercapnic challenges (termed hemodynamic response imaging (HRI)). Nineteen healthy controls and seven patients with primary brain tumors were included: six with glioblastoma (two newly diagnosed and four with recurrent tumors) and one with atypical-meningioma. Maps of percent signal intensity changes (ΔS) during hyperoxia (carbogen; 95%O2+5%CO2) and hypercapnia (95%air+5%CO2) challenges and vascular reactivity mismatch maps (VRM; voxels that responded to carbogen with reduced/absent response to CO2) were calculated. VRM values were measured in white matter (WM) and gray matter (GM) areas of healthy subjects and used as threshold values in patients. Significantly higher response to carbogen was detected in healthy subjects, compared to hypercapnia, with a GM/WM ratio of 3.8 during both challenges. In patients with newly diagnosed/treatment-naive tumors (n = 3), increased response to carbogen was detected with substantially increased VRM response (compared to threshold values) within and around the tumors. In patients with recurrent tumors, reduced/absent response during both challenges was demonstrated. An additional finding in 2 of 4 patients with recurrent glioblastoma was a negative response during carbogen, distant from tumor location, which may indicate steal effect. In conclusion, the HRI method enables the assessment of blood vessel functionality and reactivity. Reference values from healthy subjects are presented and preliminary results demonstrate the potential of this method to complement perfusion imaging for the detection and follow up of angiogenesis in patients with brain tumors.