ANNs For Predicting Prostate Cancer Recurrence

Determination of whether or not cancer will recur or progress is an important goal in prostate cancer management. Snow and coworkers reported a model for predicting recurrence23 using data from 240 men undergoing radical prostatectomy. The inputs to the ANN were age, PSA level, clinical stage, tumor grade, potency and race. The outcome of interest was cancer recurrence, characterized by biochemical (PSA) failure, local recurrence in the prostate bed or distant metastasis. The reported sensitivity was 67% and the specificity was 100% at an output cut-off of 0.5. Although four bootstrap validation and training sets were used, the validation sets were small (5% of the data base). In addition, the validation sets were used to pick the best weight matrices and, thus, the ability of this model to generalize to new patient data is questionable. Douglas and coworkers reported a sensitivity of 100% and specificity of 96% for predicting recurrence in radical prostatectomy patients using 40 clinical and pathological variables.37 Mattfeldt and associates reported a feasibility study to predict cancer progression after radical prostatectomy.38 Input variables from 40 patients included the histopathological variables: Gleason sum, World Health Organization (WHO) grade and maximum diameter of the tumor transects. Morphometric variables included volume and surface area of the epithelial tumor component and the surface area of the lumina of the neoplastic glands per unit tissue volume. An ANN using the three histopathological variables correctly predicted progression in 85% of validation cases. An ANN model using four morphometric variables correctly predicted progression in 93% of the cases.

In a study comparing ANN methods to logistic regression and Cox regression, Potter and colleagues developed models to predict progression in a selected group of patients (n = 214) at intermediate risk of progression after radical prostatectomy.39 The input variables used included age, Gleason sum, extra prostatic extension, surgical margin status, quantitative nuclear grade and DNA ploidy. The ANN models outperformed the regression models with a maximum average AUROC of 0.74 ± 0.04 for the ANN method versus maximum average AUROC of 0.68 + 0.06 for the logistic regression methods. Paired sensitivities and specificities were also significantly higher in the ANN models. The authors concluded, in part, that ANNs hold promise for predicting progression in

Artificial Neural Networks for Predictive Modeling in Prostate Cancer

Sleeping Solace

Sleeping Solace

How To Better Your Sleep For A Better Life. Understanding the importance of good sleeping habits is very beneficial to the overall health of an individual in both mental and physical levels. Learn all the tricks here.

Get My Free Ebook

Post a comment