Resumen de Automated detection and classification of brain injury lesions on structural magnetic resonance imaging
Brain injury constitutes a serious social and health problem of increasing magnitude and of great diagnostic and therapeutic complexity. Its high incidence and survival rate, after the initial critical phases, makes it a prevalent problem that needs to be addressed. According to the World Health Organization (WHO), brain injury will be among the 10 most common causes of disability by 2020. Neurorehabilitation improves both cognitive and functional deficits and increases the autonomy of brain injury patients. The incorporation of new technologies to the neurorehabilitation tries to reach a new paradigm focused on designing intensive, personalized, monitored and evidence-based treatments. These four characteristics ensure the effectivity of neurorehabilitation.
Contrary to most medical disciplines, it is not possible to link symptoms and cognitive disorder syndromes, to assist the therapist. Currently, neurorehabilitation treatments are planned considering the results obtained from a neuropsychological assessment battery, which evaluates the functional impairment of each cognitive function (memory, attention, executive functions, etc.).
The research line on which this PhD falls under aims to design and develop a cognitive dysfunctional profile integrating the results obtained in the assessment battery with information theoretical (anatomical structures and functional relations) and structural information, obtained from medical imaging studies. Therefore, it is necessary to design and develop new techniques allowing to extract structural information automatically from medical imaging studies, such as magnetic resonance imaging (MRI).
The classical approach used to identify structures consists of manually delineating brain anatomical regions. This approach presents several problems related to criterial inconsistencies of inter and intra specialist, time and repeatability. The automation of this procedure is fundamental to ensure an objective extraction of the information. Automated delineation of brain structures is usually done by registering brain images to one another or to an anatomical atlas. However, pathological changes caused by brain injury are always associated to intensity abnormalities and location alterations of the structures. This causes the traditional registration algorithms based on intensity information not to work properly and require the intervention of the clinician to select certain points on the image (which in this PhD are named singular points).
This PhD proposes a methodology to automatically detect injured anatomical structures integrating algorithms whose main objective is to generate results that may be reproducible and objective. It is divided into four stages: pre-processing, singular points identification, registration and lesion detection.
The main contributions and results obtained are: Singular points identification. This stage aims to automatize the identification of anatomical points (singular points) which clinicians have to manually identify in the event of lesions. This automation allows to identify a higher number of points by increasing the extracted information of the image and eliminating both the variability and the time invested in the manual selection of points. The main consequence is that results are reproducible and objective. This PhD proposes the design and development of a singular points detection method based on descriptors (‘Brain Feature Descriptor’ (BFD)), optimized for MRI studies. Compared to other algorithms used in the state of the art, the obtained results show that BFD allows to better characterize these medical imaging studies, regarding the average number of detected singular points, performance and point distribution in brain region.
Registration. A parametric registration algorithm based on descriptors (‘Brain Feature Registration Algorithm’ (BFRA)) is proposed. This algorithm has been compared to fourteen registration algorithms of the state of the art in three brain atlases (LPBA40, MGH10, and CUM12). BFRA improves the overlapping measures used to assess these algorithms by 17% in the case of the target overlap (TO), 6% in Dice coefficient (DC) and 11% in the Jaccard coefficient (JC). BFRA has been also compared to other rigid point set registration algorithms improving as well the aforementioned overlapping parameters. Regarding MRI studies of brain injury patients, BFRA has been compared to other widely used registration algorithm (‘Statistical Parametric Mapping’ (SPM)). The obtained results show that SPM increases the initial overlapping by 3%, whereas BFRA increases it by 15%.
Lesion identification. This final stage aims to identify those anatomical structures whose spatial location and size have been modified. A statistical study is performed to establish the normal statistical parameters associated to the location and size of the manually delineated anatomical structures in the atlas. In order to assess the quality of lesion detection, the ROC space is analyzed. The number of false negatives obtained is always zero. This means that those injured anatomical structures are always correctly detected. With regards to the value of AUC (0.79), the proposed lesion detector can be classified as acceptable.
In conclusion, this PhD corroborates the investigated research hypotheses regarding the automatic identification of lesions based on structural magnetic resonance imaging studies. Based on these foundations, new research fields to improve the automatic identification of lesions in brain injury can be proposed.
