A PERSONALIZED MICROWAVE ABLATION TREATMENT PLANNING USING RADIOMICS AND SIMULATION-BASED MODELING

TOPIC: Lung Cancer TYPE: Original Investigations PURPOSE: Minimally invasive microwave ablation (MWA) treatment is increasingly used to treat lung tumors in patients with poor lung function who cannot tolerate surgery or additional radiation therapy. Current approaches for planning ablation treatments utilize a general matrix of ablation zone dimensions, typically derived from ex-vivo non-human tissue, to provide guidance on the microwave energy dose needed to achieve local tumor control and balance safety and adequate ablation margins. Here, we present a novel method to predict ablation volume using a bioheat transfer model of microwave tissue ablation based on microwave sensitive radiomic features. METHODS: We performed a retrospective study using pre- and post-treatment CT data from 50 patients who underwent percutaneous MWA. Segmentation of the lungs, blood vessels, airway and tumor as well as radiomic analysis of the tumor tissue and surrounding region were performed on the pre-ablation CT data. A tumor region is extracted by a cascaded segmentation where an EfficientDet neural network is performed to localize the tumor and voxel-level segmentation is achieved by a U-Net. Radiomic analysis on 9 different features of the segmented tumor (Subtlety, Internal Structure, Calcification, Margin, Lobulation, Spiculation, Texture, Malignancy, and Sphericity) were conducted using Convolution Neural Networks. Segmented anatomy and radiomic analysis results were used to predict the anticipated ablation zone extent for a candidate set of applied microwave energy dose. Simulation-based prediction of the ablation extent were compared against pre-ablation tumor volume and post-ablation scar volume. Average Absolute Error (AAE) and the Dice Similarity Coefficient were used to evaluate the results of predicted simulated ablation against the manufacturer-provided geometric model (Perseon Single ST, Varian Medical Systems, Palo Alto, CA). RESULTS: Our simulation model resulted in 36% improvement in AAE and 43% improvement in Dice Similarity Coefficient compared to the manufacturer-provided geometric model. CONCLUSIONS: Combination of radiomic analysis and simulation-based modeling may better predict the extent of ablation scar after MWA and improve both safety and effectiveness of the procedure. CLINICAL IMPLICATIONS: In this retrospective study, personalized microwave ablation treatment planning provided better prediction of the actual shape of scar tissue resulting from ablation treatment, which may contribute to improvement in safety and local tumor control. DISCLOSURES: No relevant relationships by Faraz Chamani, source=Web Response No relevant relationships by Damian Dupuy, source=Web Response No relevant relationships by Natthapat Khuwijitjaru, source=Web Response No relevant relationships by Natthawat Khuwijitjaru, source=Web Response PI ofan investigator-initiated study relationship with Medtronic Please note: >$100000 by Fabien Maldonado, source=Web Response, value=Grant/Research Supportco-I on investigator-initiated studies relationship with Janssen Please note: $20001 - $100000 by Fabien Maldonado, source=Web Response, value=Grant/Research Support Removed 04/19/2021 by Fabien Maldonado, source=Web ResponsePI on investigator-initiated relationship with Erbe Please note: $5001 - $20000 by Fabien Maldonado, source=Web Response, value=Grant/Research Support Consulting relationship with Medtronic Please note: $5001 - $20000 by Fabien Maldonado, source=Web Response, value=Honorariaco-I industry-sponsored trial relationship with Lung Therapeutics Please note: $5001 - $20000 by Fabien Maldonado, source=Web Response, value=Grant/Research Support speaker relationship with Longitudinal lung nodule fellows program - Please note: $1001 - $5000 by Fabien Maldonado, source=Web Response, value=Honoraria Removed 04/19/2021 by Fabien Maldonado, source=Web Response Board of director member relationship with AABIP Please note: $1-$1000 by Fabien Maldonado, source=Web Response, value=Travel Employee relationship with phenoMapper Inc. Please note: 2020 - Present Added 04/30/2021 by Chanok Pathompatai, source=Web Response, value=Salary No relevant relationships by Punit Prakash, source=Web Response No relevant relationships by Jan Sebek, source=Web Response Consultant relationship with phenoMapper Inc. Please note: Oct 2019-Present Added 04/29/2021 by Pinyo Taeprasartsit, source=Web Response, value=Grant/Research Support Owner/Founder relationship with phenoMapper, LLC Please note: 01/2021-12/2021 Added 04/30/2021 by Henky Wibowo, source=Web Response, value=Grant/Research Support