May 2003
Volume 44, Issue 13
Free
ARVO Annual Meeting Abstract  |   May 2003
Prediction of Metastatic Death From Uveal Melanoma Using a Bayesian Artificial Neural Network
Author Affiliations & Notes
  • B.E. Damato
    St Paul's Eye Unit, Royal Liverpool Univ Hospital, Liverpool, United Kingdom
  • A.C. Fisher
    Dept of Clinical Engineering, Royal Liverpool Univ Hospital, Liverpool, United Kingdom
  • A.F. Taktak
    Dept of Clinical Engineering, Royal Liverpool Univ Hospital, Liverpool, United Kingdom
  • Footnotes
    Commercial Relationships  B.E. Damato, None; A.C. Fisher, None; A.F.G. Taktak, None.
Investigative Ophthalmology & Visual Science May 2003, Vol.44, 2159. doi:
  • Views
  • Share
  • Tools
    • Alerts
      ×
      This feature is available to authenticated users only.
      Sign In or Create an Account ×
    • Get Citation

      B.E. Damato, A.C. Fisher, A.F. Taktak; Prediction of Metastatic Death From Uveal Melanoma Using a Bayesian Artificial Neural Network . Invest. Ophthalmol. Vis. Sci. 2003;44(13):2159.

      Download citation file:


      © ARVO (1962-2015); The Authors (2016-present)

      ×
  • Supplements
Abstract

Abstract: : Purpose: To create an artificial neural network expert system with a probabilistic algorithm based on Bayes’ theorem for predicting survival in patients with uveal melanoma. Methods: The survival time of 2543 patients with uveal melanoma was divided into 5 intervals, each containing an equal number of events. After date censoring, records were divided into training and test sets, according to odd or even birth year respectively. The expert system was tested against Cox’s regression and Kaplan Meier analysis. The artificial expert system was also assessed against a clinical expert in a double-blind trial prediction of 30 previously unseen records. Results: Five predictive factors were identified using an iterative ROC analysis: coronal tumor location; sagittal tumor location; anterior tumor margin; largest basal tumor diameter; and cell type. These were used as input covariates for the final expert system. The RMS error for survival prediction by the expert system was 3.8 years compared with 4.3 years by the clinical expert. The expert system performed better in patients with poor prognosis (survival time < 4 years); conversely, predictions by the clinical expert were more accurate for patients with good prognosis. In approximately 40% of cases Kaplan Meier analysis was not possible due to an insufficient number of records with similar input parameters in the database. Conclusions: This analysis identifies the most significant input covariates in the prediction of survival from uveal melanoma. Our expert system using a Bayesian probability model combined with an artificial neural network is at least as accurate as the clinical expert. This method should be useful in predicting outcomes for rare conditions, if Kaplan-Meier analysis is unsuitable and if clinical expert opinion is unavailable.

Keywords: melanoma • tumors • oncology 
×
×

This PDF is available to Subscribers Only

Sign in or purchase a subscription to access this content. ×

You must be signed into an individual account to use this feature.

×