Metodologías de cuantificación y clasificación de imágenes de medicina nuclear para la ayuda a la decisión clínica
- Autores: A.P. Seiffert
- Directores de la Tesis: Luis Enrique Gómez Aguilera (dir. tes.), Patricia Sánchez González (codir. tes.)
- Lectura: En la Universidad Politécnica de Madrid ( España ) en 2021
- Idioma: español
- Materias:
- Texto completo no disponible (Saber más ...)
- Resumen
Neurodegenerative diseases present a significant and increasing burden on the population. They are a group of diseases characterised by the progressive degen-eration or death of neurons, resulting in reduced brain perfusion and metabolism. Dementia is a syndrome caused by a neurodegenerative disease and is characterised by cognitive decline. Neurodegenerative diseases such as Alzheimer’s disease (AD) are also characterised by abnormal aggregation of proteins in the brain that can be present even before the onset of clinical symptoms. AD presents two types of protein deposition: (i) extracellular neuritic amyloid-beta (Aβ) plaques and (ii) intracellular hyperphosphorylated tau.
Positron emission tomography (PET) allows in vivo evaluation of these AD biomarkers. On one hand, [18F]Fluorodeoxyglucose (FDG) PET visualises the brain metabolism, with different neurodegenerative diseases presenting characteristic pat-terns of reduced cortical activity. On the other hand, specific Aβ-binding radiotrac-ers like [18F]Florbetapir (FBP), [18F]Florbetaben (FBB), and [18F]Flutemetamol (FMM) show increased Aβ deposition in the brain cortex. Therefore, PET imaging is an im-portant tool in clinical routine practice in the diagnosis of neurodegenerative diseas-es. While usually the images are interpreted visually, methodologies for the quantifi-cation and classification are required to improve the clinical decision-making.
In this PhD thesis, a new methodology is proposed for the processing and analysis of PET neuroimages, for creating clinical decision support systems. The pro-posed methodology is validated by studying its application in three specific clinical contexts, defining research hypotheses in each case: (i) the evaluation of motion cor-rection (MoCo) of Aβ PET images and its impact on diagnosis, (ii) the texture analy-sis of Aβ PET images and its classification capabilities, and (iii) the quantitative vali-dation of the perfusion-like PET data of the first-minute-frame (FMF) after 18F-labelled Aβ-binding radiotracer injection compared to [18F]FDG PET.
The proposed methodology allows the preprocessing and quantification of PET neuroimages, and the data is statistically analysed. The following results of the PhD thesis can be grouped according to the clinical problems the methodology is applied to. First, the Aβ deposition of MoCo and non-MoCo [18F]FBB PET images is quantified by applying the designed image processing methodology. Slight changes to the regional quantitative values can be observed after MoCo (generally < 0.03). Additionally, to evaluate the impact in clinical routine practice, MoCo and non-MoCo images are visually interpreted, showing high diagnostic agreement (generally over 90%) but an increase of diagnostic confidence after MoCo, especially in patients with large head motion.
Regarding the texture analysis of Aβ PET images, the extracted texture fea-tures are proposed to measure the grey-to-white matter contrast, which is reduced or absent in patients with high Aβ load. Depending on the cortical uptake, images are classified into Aβ positive and Aβ negative in visual interpretation. Texture features are extracted from 66 Aβ PET images acquired with [18F]FBP or [18F]FMM. All 6 ex-tracted features show good discriminatory performance with area under the curve values of up to 0.949 when classifying into Aβ positive and Aβ negative. Additional-ly, classifiers based on the support vector machine algorithm were created reaching 100% accuracy.
The FMF is presented as an alternative to [18F]FDG PET imaging for the eval-uation of neuronal injury. It provides a perfusion-like image that can be acquired with the same radiotracer dose as Aβ PET images. In a database of 60 cognitively impaired patients, the quantitative similarity between the FMF acquired with [18F]FBP, [18F]FBB or [18F]FMM, and [18F]FDG PET images is investigated. Linear cor-relation analyses of regional quantitative radiotracer uptake measures result in high correlation coefficients. Concretely, the mean intrapatient correlation coefficient is r = 0.93 ± 0.05, and regional interpatient correlation coefficients reach r = 0.92. There-fore, the comparability of both images is quantitatively demonstrated.
Lastly, the application of the designed methodology is not limited to these three clinical problems. Results are successfully obtained evaluating the prognostic value of the Aβ deposition in the striatum, quantitatively comparing [18F]FDG PET to perfusion imaging by arterial spin labelling magnetic resonance imaging, and detect-ing patterns in [18F]FDG PET images of patients with primary progressive aphasia, and Parkinson’s disease and Parkinsonisms.
The results obtained in this PhD thesis confirm the investigated research hy-potheses and application of PET neuroimage quantification and classification meth-odologies in three relevant clinical problems. Future research lines are proposed re-garding the investigated clinical problems, and to apply the designed methodology to other clinical problems using neuroimaging, and even in other contexts not related directly to neurology.
