PREDICTIVE SYSTEMS IN ONCOLOGY BASED ON REAL LIFE PATIENTS RECORDS
Piotr Stępniak , Konrad Wojdan , Joanna Chodzyńska , Lucjan S Wyrwicz , Konrad Świrski
AbstractAnonymous records of patients from Maria Sklodowska-Curie Memorial Cancer Center were obtained and analyzed for usefulness in predictive algorithms. Classification and approximation of patient condition were attempted using Support Vector Machines and Random Forests. Analysis of initial data faced many obstacles including incompleteness of records and various errors. Prepared predictive models were based on blood test parameters for given oncological diagnosis and proposed treatment. They predicted if the patient was likely to suffer from complications after surgery by estimating the number of hospital stay days. Best model was able to identify 28.6% of risk patients with 0 false positives in validation. The studies resulted in preparation of a prototype system to support oncological decisions. Even though obtained sensibility is not very impressive for contemporary AI methods it might mean saving one of four people from suffering or death. The prepared methodology enables achieving better results with high quality data.
|Publication size in sheets||0.5|
|Book||Palma dos Reis António (eds.), Ajith P. Abraham Aijth: Proceedings of the IADIS International Conference e-Learning 2013, 2013, ISBN 978-972-8939-88-5|
* presented citation count is obtained through Internet information analysis and it is close to the number calculated by the Publish or Perish system.