Interpretation of laboratory results and elaboration of a clinical diagnosis in blood coagulation domain
Résumé
During a haemorrhage, a phenomenon called coagulation stops the blood effusion. It can be measured with coagulation curves known as thrombogram. In some individuals affected with haemophilia this phenomenon is disturbed. If some precautions are not taken their life could be put in danger while prescribing a treatment or performing a surgery. Thus it is important for clinicians to establish an accurate diagnosis. However, providing such a diagnosis is a complicated task as a lot of parameters have to be taken into account. The goal of this study is the realisation of a clinical decision support system that would help health professionals during the process of decision making. Machine learning techniques are used to classify thrombograms and therefore, to detect haemophilia, its type, A or B, and its severity. This paper presents a comparison of different classification methods and the selection criteria identified to choose the best method. On a dataset of 14 000 artificially generated thrombograms, results show 98\% recognition for haemophilia and 93\% recognition for the different types and severities of haemophilia. It proves that our system is a viable solution and that machine learning techniques can efficiently detect haemophilia and provide a reliable complete diagnosis.