IMG-11. A COMPUTERISED CLINICAL DECISION SUPPORT SYSTEM FOR DIAGNOSING CHILDREN'S BRAIN TUMOURS USING FUNCTIONAL IMAGING AND MACHINE LEARNING. (3rd June 2022)

• Record Type: Journal Article
• Authors: Rose, Heather; Ahmed, Arfan; Babourina-Brooks, Ben; Khan, Omar; MacPherson, Lesley; Manias, Karen; Peake, Ashley; Ali, Sana; Withey, Stephanie; Worthington, Lara; Novak, Jan; Zarinabad, Nilou; Grundy, Richard; Arvanitis, Theodoros; Peet, Andrew
• Is Part Of: Neuro-oncology. Volume 24(2022)Supplement 1
• Page Start: i79
• Page End: i79
• Publication Date: 2022-06-03
• Subjects: Brain Neoplasms -- Periodicals; Brain -- Tumors -- Periodicals; Brain -- Cancer -- Periodicals; Nervous system -- Cancer -- Periodicals; 616.99481
• DOI: 10.1093/neuonc/noac079.287
• Languages: English
• ISSNs: 1522-8517

Abstract

Abstract: INTRODUCTION: Magnetic resonance imaging is a key investigation in the diagnosis of childhood solid tumours. Advanced techniques such as diffusion weighted imaging (DWI), magnetic resonance spectroscopy (MRS) and perfusion imaging probe the underlying cellular, chemical and vascular nature of the disease. Coupled with machine learning these scanning methods show improvement in diagnostic accuracy compared with conventional imaging. Advanced image analysis is not routinely available in hospitals. We present a clinical decision support system (CDSS) developed for advanced MR analysis and interpretation. METHOD: The CDSS was developed in house. The Children's Cancer and Leukaemia Group Functional Imaging Group (CCLGFIG) Database, a national resource, was used to provide a repository of cases together with their advanced imaging and machine learning diagnostic classifiers. A new case is displayed alongside cases in the repository with known diagnoses, including summary statistics for relevant diagnostic categories. The CDSS was made available to radiologists, in their clinical environment for technical and clinical evaluation. Structured interviews were undertaken. The CDSS was developed as a computer app for multi-centre distribution. RESULTS: 436 MRS, 240 DWI and 85 perfusion cases were available for building repositories. Machine learning classifiers showed diagnostic accuracies for the major childhood brain tumour types of 85-95%. Comparison of MRS with a data repository was found to improve non-invasive diagnosis. Results from the CDSS can be uploaded to the CCLGFIG to support multicentre research. Positive feedback on the CDSS from clinicians included: ready access to advanced analysis; simple and efficient integration into clinical workflow; and assisted interpretation of advanced analysis. DISCUSSION: Advanced MR analysis techniques provide improved non-invasive diagnostic accuracy but are difficult to implement on clinical systems due to technical, infrastructure and training limitations. CONCLUSION: We have successfully released a CDSS for paediatric cancer within the hospital environment and assessed its suitability for clinical use.