P09.62 Towards individualized survival prediction in glioblastoma patients using machine learning methods

Abstract

INTRODUCTION: Several recent retrospective analyses have shown conflicting correlation of doses of radiotherapy (RT) to SVZ with survival in patients with glioblastoma (GBM). We present the results of a prospective study evaluating RT doses received by SVZ in patients diagnosed with newly-diagnosed GBM and their impact on survival. MATERIAL AND METHODS: Between 2012-2016, 100 patients with newly diagnosed GBM were accrued in this IRB approved prospective study. 80 patients who completed their adjuvant treatment form the study group for the present analysis. Patients underwent maximal safe surgical resection followed by adjuvant focal conformal RT to a dose of 59.4 Gy/33 fractions with standard concurrent temozolomide and 6-12 adjuvant temozolomide chemotherapy. We did not specifically consider SVZ during RT planning, which was done as per standard criteria of tumour bed+2-3 cms margin as CTV+5 mm margin as PTV. Tumours were divided into groups based on their spatial relationship to SVZ. Ipsilateral, contralateral, and bilateral SVZs were contoured on RT treatment plans by the use of co-registered MRI and CT scans. Multivariate Cox regression was used to examine the relationship between SVZ dose and progression-free survival (PFS) and overall survival (OS). PFS was measured by the RANO criteria and MR Spectroscopy and MR perfusion were used routinely. RESULTS: Median age of patients was 49 years (17-78) and median KPS was 80. GTR was achieved in 22 patients (27.5%), STR in 47 patients (58.7%) and biopsy in 11 patients (13.8%). Median PFS and OS of the entire cohort of 80 studied patients were 13.43 months and 15.4 months respectively. The grouping of SVZ showed 29 patients (36%) in group I, 14 patients (18%) in group II, 29 patients (36%) in group III, and 8 (%) patients in group IV, which however did not influence PFS and OS significantly. The mean and median doses received by the ipsilateral SVZ was 54.2 Gy and 59 Gy (0-60.99 Gy), for contralateral SVZ was 43.7 Gy and 47.84 Gy (0-60.88 Gy) and for combined SVZ was 48.8 Gy and 50.87 Gy (12.12-60.46), respectively. On multivariate analysis, patients who received <59 Gy to the ipsilateral SVZ had significantly better outcomes than patients who received > 59 Gy to the the ipsilateral SVZ both in terms of median PFS (20.5 vs. 9.7 months, p=0.016) and OS (20.6 vs. 13.3 months, p=0.026). Patients who received <47.85 Gy to the contralateral SVZ had significant longer median PFS than those who received >47.85 Gy (35 vs 10.9 months, p=0.014) but had no impact on OS. MGMT was done in 45 patients, 18 patients were methylated with median OS of 16.8 months. CONCLUSION: Our trial, probably the first prospective study correlating the relationship of SVZ dosimetry with survival in a reasonably large number of patients demonstrates an inverse correlation between the RT doses received by ipsilateral and contralateral SVZ with survival, and should set a benchmark to plan appropriate studies.

A three-dimensional computational analysis of magnetic resonance images characterizes the biological aggressiveness in malignant brain tumours

“…An indirect confirmation of the prediction was reported in [12]. Several studies have used machine-learning approaches to construct prognosis models from sets of descriptors of tumour shape, texture, etc., obtained from different MRI sequences [15][16][17][18][19][20]. Finally, other authors have explored different types of spatial imaging biomarkers based on the quantification of tumour subregions [21][22][23].…”