During the last decades, the availability of great amounts of biomedical information has fostered the development of tools to allow the extraction and inference of knowledge. The increase of biomedical technologies to assist the clinical experts in their decisions has contributed to the incorporation of an evidence-based medicine paradigm centered in the patient.
The contributions of this Thesis are focused in the development of tools to assist in the clinician's workflow decision process in the diagnosis of brain tumours (BTs) using  Magnetic Resonance Spectroscopy (MRS).
This Thesis contributes with the development of Pattern Recognition (PR)-based classifiers trained with MRS data and devoted to child and adult patients for tumour type and aggressiveness level assessment. These classifiers take advantage of the biochemical differences found in BT in children and adults in order to provide discrimination.
The development of classification models aimed at the discrimination of the three most prevalent pediatric brain tumour types  is one of the major contributions of this Thesis. A common location of these tumours is the cerebellum, where it is difficult to distinguish the tumour type with Magnetic Resonance Imaging alone. Hence, obtaining high accuracy in the discrimination from MRS data of pilocytic astrocytomas, ependymomas and medulloblastomas is crucial to stablish a surgical strategy for tumour resection, since each tumour type requires different actions to be taken to obtain good prognosis. 
In addition, it is concluded that the combination of single voxel MRS at 1.5T at two different Echo Time (TE), Short-TE and Long-TE, improves the classification of pediatric brain tumours over the use of one TE alone. This finding extends and corroborates similar results achieved with MRS data from adults. 
A novel on-line method to audit predictive models using a Bayesian perspective for Decision Support Systems (DSSs) devoted to clinical environment is also presented in this Thesis. This audit method positively affects and improves the clinician's decision workflow in a clinical environment by deciding which is the classifier that best suits each particular case being evaluated and by allowing the detection of possible misbehaviours due to population differences or data shifts in the clinical site. The efficacy of such a method is shown for the problem of diagnosis with a multi-centre database of MRS data of BTs. 
This Thesis complements the audit method by contributing with a methodology for prior probability assessment to a set of classifiers. The similarity model, inspired also in the Bayesian approach, allows the DSS to select the most adequate classifier for each test case attending to contextual information, which is information not used in the design of the classifiers but related to the case or its environment.
The results of this Thesis have directly contributed to the eTUMOUR (Web accessible MR decision support system for brain tumour diagnosis and prognosis, incorporating in vivo and ex vivo genomic and metabolomic data, 2004-2009), and HEALTHAGENTS (Agent-based Distributed Decision Support System for Brain Tumour Diagnosis and Prognosis, 2006-2008) European Union projects of the 6th Framework Programme. As a result to the scientific contributions studied in the Thesis, two traslational applications can be emphasized. They consist in the incorporation of  practical solutions to improve the clinical decision workflow supplied by CURIAM BT, a clinical DSS for BT diagnosis support: the incorporation of the pediatric classifiers as an effective non-invasive pre-operative tool to define the tumour resection strategy; and the incorporation of the audit method and of the similarity model as tools to select the adequate classifier for each case.