Meeting Banner
Abstract #1211

Textural Analysis of Echo-Planar Diffusion-Weighted Imaging Improves Preoperative Characterisation of Suspected Thyroid Tumours

Anna M. Brown1, 2, Sidhartha Nagala3, Daniel Scoffings4, Mary McLean2, Andrew N. Priest4, Piyush Jani3, John Griffiths2

1School of Medicine, Duke University, Durham, NC, United States; 2Department of Oncology, University of Cambridge, Cambridge, Cambridgeshire, United Kingdom; 3Department of Otolaryngology, Addenbrookes Hospital, University of Cambridge, Cambridge, Cambridgeshire, United Kingdom; 4Department of Radiology, Addenbrooke's Hospital, University of Cambridge, Cambridge, Cambridgeshire, United Kingdom

Ultrasound-guided needle biopsy fails to diagnose up to 20% of malignant thyroid nodules. Texture analysis and linear discriminant analysis of apparent diffusion coefficient maps offers a novel approach to non-invasively distinguish between benign and malignant thyroid nodules. We achieved high sensitivity (87%) and specificity (100%) on a training set, with 83/87 slices and 18/20 nodules correctly classified. Additionally, 11/14 slices and 5/6 nodules were correctly characterised in a test set. This study indicates great potential to better characterise indeterminate thyroid nodules, and a large prospective study is now needed to fully prove this model.

Abstract PDF Poster PDF

Keywords

account accuracy achieved achieving apparent applications around assess audience basis benign best biomedical blinded brown calculation cancer charitable classification classified classifies clinical coefficient comparing complex consistent consultant correctly curve curves definitive diagnostic differentiate differentiation diffusion discovered discriminant discriminating discrimination discriminatory distinguished divided drew dual duke efficacy experimental extracellular fail feature field fully gave generated giving great held highest hospital hypothesis identified improve improves indicate isolating john kingdom lesions like linear malignant matrix medicine model need needed nodule nodules oncology operations output pathology patient patients planar potential preoperative priest prism procedure promising prospective prove radiologist radiologists radiology reasonably recent reduction reliably reports represented researchers restricted resultant school scoring selected selecting sensitivity sets since slice slices space spacing spared specificity spin supports suspected target technical textural texture thank thyroid tool training trust tumor unnecessary view whether