NIMG-07. APPLYING A GLIOMA-TRAINED DEEP LEARNING AUTO-SEGMENTATION TOOL ON BM PRE- AND POST-RADIOSURGERY

Abstract PURPOSE Stereotactic radiosurgery (SRS) has become the mainstay to treat BM. Follow-up MRI provides important information on lesion treatment response and guides future therapy planning. Volumetric measurements of BM have shown promise over traditional uni- two-dimensional in more accurate repeatable assessment. However, routine clinical use yet be achieved because workflow is laborious. In previous work, we developed a PACS-integrated deep learning algorithm for automatic high- low-grade glioma 3D segmentation. this applied U-Net segment pre- post-Gamma Knife (GK) evaluated performance. METHODS 10 post-GK studies were autosegmented five randomly selected patients (melanoma n= 3, breast 2). The trained segmented “Whole Tumor” (tumor core+peritumoral edema T2w-FLAIR) “Tumor Core” (CE tumor core+necrosis SPGR). AI generated segmentation was then revised as needed by board-certified neuroradiologist dice-similarity-coefficient (DSC) between volumetric segmentations calculated. RESULTS Four had multicentric (2-4 BM) lesions. mean± SD DSC Whole Tumor Core 0.92±0.06 0.46±0.30 pretreatment, 0.84±0.09 0.41±0.25 posttreatment BM, respectively. tool detected lesions with sensitivity 45% (5/11) pretreatment 50% (3/6) Three all that not autosegmentation showed very faint hyperintense peritumoral T2w-FLAIR. CONCLUSION FLAIR using glioma-trained did require major adjustment if it detects lesion. On other hand, low detection enhancing component, dedicated training annotated data will needed.