NIMG-65. IMPROVED SPATIAL MAPPING OF TUMOR AGGRESSIVENESS WITH 1H MAGNETIC RESONANCE SPECTROSCOPY AND DEEP LEARNING IN PATIENTS WITH NEWLY-DIAGNOSED GLIOMA

Abstract INTRODUCTION Noninvasive, radiopathomic mapping of tumor aggressiveness can benefit patients with glioma by guiding the selection tissue samples for diagnosis, increasing extent resection, and non-invasively characterizing residual burden subsequent treatment. Although prior studies have demonstrated utility metabolic metrics quantified from 1H-MR Spectroscopy (MRS) in probing pathology, this study evaluated using entire 1D-spectrum deep learning intratumoral cellularity, proliferation (ki-67), a new index (TAI) defined as log((n(ki−67)+n(cellularity))*tumor-score). METHODS Multi-voxel 1H-MRS was acquired on 281 newly diagnosed (47% IDH-wildtype) immediately before surgical resection. After reconstructing individual spectra at locations where were obtained during surgery normalizing NAA contralateral normal-appearing-white-matter, 607 corresponding histopathology deemed sufficient quality analysis. A 1D convolutional-neural-network bidirectional long- short-term memory deep-learning model spectrum (0.6-3.6ppm) compared to mixed-effects regression (with choline-to-NAA index[CNI]) Random Forest CNI+normalized peak heights) models predicting ki-67, TAI. Results & DISCUSSION Using resulted 10.3%-22.1% lower mean absolute error (MAE) 0.32-0.37 higher R2 values CNI alone or random forest multiple metrics. MAE all 3 26-44% < 1 standard deviation ground truth, demonstrating reasonable prediction accuracy within test data set. lowest (0.16) highest (0.41) attained when TAI deep-learning, cellularity %MAE. Colormaps predicted pathology identified regions heightened surrounding most abnormal pathological features that sometimes extended beyond non-enhancing lesion. Current work is evaluating clinical our maps aggressiveness.