Artificial intelligence for personalized management of vestibular schwannoma: A clinical implementation study within a multidisciplinary decision making environment

BackgroundThe management of patients with Vestibular Schwannoma (VS) relies heavily on precise measurements of tumour size and determining growth trends. MethodsIn this study, we introduce a novel computer-assisted approach designed to aid clinical decision-making during Multidisciplinary Meetings (MDM) for patients with VS through the provision of automatically generated tumour volume and standard linear measurements. We conducted two simulated MDMs with the same 50 patients evaluated in both cases to compare our proposed approach against the standard process, focusing on its impact on preparation time and decision-making. FindingsAutomated reports provided acceptable information in 72% of cases, as assessed by an expert neuroradiologist, while the remaining 28% required some revision with manual feature extraction. The segmentation models used in this report generation task achieved Dice scores of 0.9392 ({+/-} 0.0351) for contrast-enhanced T1 and 0.9331 ({+/-} 0.0354) for T2 MRI in delineating whole tumor regions. The automated computer-assisted reports that included additional tumour information initially extended the neuro-radiologists preparation time for the MDM (2m 54s ({+/-} 1m and 22s) per case) compared to the standard preparation time (2m 36s ({+/-} 1m and 5s) per case). However, the computer-assisted simulated MDM (CAS-MDM) approach significantly improved MDM efficiency, with shorter discussion times per patient (1m 15s ({+/-} 0m and 28s) per case) compared to standard simulated MDM (SS-MDM) (1m 21s ({+/-} 0m and 44s) per case). InterpretationThis pilot clinical implementation study highlights the potential benefits of integrating automated measurements into clinical decision-making for VS management. An initial learning curve in interpreting new data measurements is quickly mastered and the enhanced communication of growth patterns and more comprehensive assessments ultimately provides clinicians with the tools to offer patients more personalized care. FundingN. Wijethilake was supported by the UK Medical Research Council [MR/N013700/1] and the Kings College London MRC Doctoral Training Partnership in Biomedical Sciences. This work was supported by core funding from the Wellcome Trust (203148/Z/16/Z) and EPSRC (NS/A000049/1) and an MRC project grant (MC/PC/180520). TV is also supported by a Medtronic/Royal Academy of Engineering Research Chair (RCSRF1819/7/34). Graphical Abstract O_FIG O_LINKSMALLFIG WIDTH=200 HEIGHT=101 SRC="FIGDIR/small/23298685v1_ufig1.gif" ALT="Figure 1"> View larger version (46K): org.highwire.dtl.DTLVardef@8f3e10org.highwire.dtl.DTLVardef@339908org.highwire.dtl.DTLVardef@1840f9corg.highwire.dtl.DTLVardef@11d0810_HPS_FORMAT_FIGEXP M_FIG C_FIG HighlightsO_LIThe first study to evaluate the impact of AI assisted reporting in a clinical setting. C_LIO_LIAI generated segmentations can be used to provide a clinical guideline driven report facilitating personalized patient management C_LIO_LIVolumetric tumour measurements provide a more comprehensive assessment of tumour growth. C_LI