Advanced optimization METhods for automated central veIn Sign detection in multiple sclerosis from magneTic resonAnce imaging

Multiple sclerosis (MS) is a chronic disease that affects the central nervous system, showing heterogeneous clinical manifestations such as motor, cerebellar and sensory symptoms. Over the past two decades, magnetic resonance imaging (MRI) has become an indispensable tool for diagnosing MS. However, the current MRI criteria for MS diagnosis have imperfect specificity [GDR12,SRT07], making misdiagnosis of MS relatively common, with relevant health and socioeconomic costs [SKB12,SBC16]. To distinguish MS lesions from white matter abnormalities arising from other diseases, the identification of an MRI-detectable vein traversing the center of a lesion has been proposed as a diagnostic tool [F64,TDD11,MAG18,SSH16]. This marker is referred to as the central vein sign (CVS). However, the clinical application of the CVS as a biomarker is limited by interrater differences in the adjudication of the CVS, as well as the time burden required for the manual determination of the CVS for each lesion in a patient’s full MR imaging scan. As of today, only two automated algorithms for the classification of MS lesions as CVS+ (MS lesions showing the presence of the CVS) or CVS- (MS lesions without the CVS) have been proposed in the literature [MFJ20,DSS18]. Nevertheless these approaches suffer from some weak points. The most evident ones are now mentioned. In [MFJ20] the starting image segmentation and registration phases are manually performed and the software is not publicly available. On the other hand, the procedure described in [DSS18] has been proved to be less accurate than manual rating in assessing all the lesions, it needs to manually evaluate the noise and to exclude high-level noisy acquisitions. The aim of this project is to develop an automated pipeline for the CVS detection able to overcome the limitations of the existing approaches, easy to be exploited in the clinical practice and supported by a strong theoretical analysis of the mathematical aspects which model all the pipeline elements needed to provide the final output. Starting from the MRI acquisitions, the CVS assessment requires addressing different imaging tasks that can be mathematically formalized as optimization problems, namely the segmentation of brain lesions, the registration of images acquired at different magnetic fields, the removal of noise and artifacts from the data and the classification of a lesion as CVS+ or CVS-. The project has the ambition to propose ad hoc numerical optimization algorithms to face the above mentioned problems and combine them in a unified pipeline suitable for fast and automatic processing of the acquired MRI brain images to identify the CVS. The automated detection of the CVS would ease its implementation as a diagnostic biomarker in clinical practice with meaningful advantages for both clinicians and patients. The ultimate goal is to apply the developed CVS detection tool to concrete MRI data provided by the University of Firenze research unit. References [DSS18] Dworkin, J.D., Sati, P., Solomon, A., et al. Am. J. Neuroradiol., 39(10) (2018). [F64] Fog T. Acta Neurol. Scand. Suppl., 40 (1964). [GDR12] Gómez-Moreno, M., Díaz-Sánchez, M., Ramos-González, A., Mult. Scler., 18 (2012). [MAG18] Maggi P., Absinta M., Grammatico M, et al., Ann. Neurol. 83(2) (2018). [MFJ20] Maggi, P., Fartaria, M.J., Jorge, J., et al., NMR Biomed., 3 (2020). [SBC16] Solomon AJ, Bourdette DN, Cross AH, et al. Neurology, 87 (2016). [SKB12] Solomon AJ, Klein EP, Bourdette D., Neurology, 78 (2012). [SSH16] Solomon AJ, Schindler MK, Howard DB, et al. Ann Clin Transl Neurol., 3(2) (2016). [SRT07] Swanton, J. K., Rovira, A., Tintore, M., et al., Lancet Neurol., 6 (2007). [TDD11] Tallantyre EC, Dixon JE, Donaldson I, et al., Neurology, 76 (2011).