Use this URL to cite or link to this record in EThOS:
Title: Neutrophil count prediction in childhood cancer patients receiving 6-mercaptopurine chemotherapy treatment
Author: Avdic, Dalila
ISNI:       0000 0004 8501 5323
Awarding Body: University of Lincoln
Current Institution: University of Lincoln
Date of Award: 2018
Availability of Full Text:
Access from EThOS:
Access from Institution:
Acute Lymphoblastic Leukaemia (ALL) is a common form of blood cancer, usually affecting children under 15 years of age. Chemotherapy treatment for ALL is delivered in three phases viz. induction (to achieve initial remission), intensification (to kill the majority of abnormal cells), and finally, maintenance. The maintenance phase involves oral administration of the chemotherapy drug 6-Mercaptopurine (6-MP) in varying doses to destroy any remaining abnormal cells and prevent reoccurrence. A key side effect of the treatment is a reduction in neutrophil counts that can result in a condition known as neutropenia, i.e. reduced immune system. This carries a risk of secondary infection and has been linked to 60% of ALL fatalities. Current practice aims to control neutrophil counts by varying 6-MP dosages on a weekly basis based on blood counts. However, its success is varied. This thesis proposes a number of intelligent prediction methods to more accurately predicting neutrophil counts one week ahead using blood count data and corresponding 6-MP dosing regimens. Firstly, a well-known and robust neural network (Nonlinear Autoregressive Exogenous) is applied to blood count data to provide an initial assessment of the feasibility of such an approach. A comparative analysis of a series of more complex algorithms is then considered for more advanced, in-depth analysis viz. Multi-Layer Perceptron (MLP) and Support Vector Machines (SVM). Both methods are shown to have a prediction accuracy of around 60% on the first sample period, with the MLP also having a prediction accuracy of more than 60% in the second sample period in seven out of ten blood data points (there was 10 timeseries blood data predictions). However, in comparison the accuracy of SVM is relatively low. Finally, an incremental learning-based approach is proposed to increase the accuracy of the system and provide a realistic framework for real-time implementation. The accuracy is shown to improve considerably as more data is added, and the predicted neutrophils data is shown to follow the trend of the actual neutrophil counts.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available