Prostate cancer is the most widespread type of cancer affecting men and the second most common cause of terminal cancer in the
Despite the prevalence of the disease, very little is known about its direct causes and deciding on a disease management plan for a patient can be difficult.
To address this, a research project being undertaken by the
On diagnosis of prostate cancer, recovery greatly depends on the stage of the disease and the health status of the patient. Therefore, early diagnosis and an appropriate management plan taking into account the health of the patient are vital to both survival and quality of life.
Diagnosed patients usually undergo further investigation to decide on the most appropriate treatment based on information currently available. Treatment has many possible side effects, impotence for example, and because surgery is not always successful a balance has to be struck between management of disease and quality of life. The treatment plan is developed by the patient and physician together, and by allowing the patient to explore the likelihood of various outcomes, will assist them in making these difficult choices.
Dr John McCall, Director of the Computational Intelligence Group at
‘It is this uncertainty that makes prostate cancer a suitable area for artificial intelligence medical decision support.
‘My team has developed a software tool to assist doctors in deciding on the best management plan for a patient.’
The software, designed in collaboration with Dr Eng Ong of the Department of Urology at Aberdeen University, uses an intelligent computer program based on the use of Bayesian Networks to discover hidden relationships between diseases and symptoms by analysing data collected during the treatment of patients.
The Bayesian Network contains data from a large population of patients who have already been treated in a variety of ways depending on their condition and captures the best course of treatment as a set of probabilities.
The software tool may be used throughout the patient journey from testing and diagnosis, to treatment and recovery or death. At many points on this journey decisions have to be made that divert the patient down one of many possible pathways. Physicians can use the software to pose questions to determine the most likely outcome to any given treatment scenario.
Practitioners currently use Partin tables, a technique allowing doctors to predict the definitive pathological stage and best course of treatment for the disease. The new software model is different, however, as it can continually learn and adapt given new information and, because its probability network is flexible, can ask a range of different questions.
The information contained in the software model is not based on a fixed set of data from a particular date or geographical location as Partin tables are, so can give accurate probabilities for the population in any part of the world. It also has the potential to be used for other purposes, including healthcare planning.
Ratiba Kabli, a Robert Gordon University research student working on the project, said: ‘The model developed can answer various queries relating to prostate cancer management from diagnosis to treatment decision making and in some cases will save patients the discomfort of undergoing invasive surgery.’
The research team tested their model on 320 patient cases at Aberdeen Royal Infirmary and initial findings compared with Partin tables showed promising results.
The next stage is to test the model on a larger scale.