Limits...
Classification models for early detection of prostate cancer.

Wichard JD, Cammann H, Stephan C, Tolxdorff T - J. Biomed. Biotechnol. (2008)

Bottom Line: We build ensembles of classification models in order to increase the classification performance.We measure the performance of our models in an extensive cross-validation procedure and compare different classification models.The datasets come from clinical examinations and some of the classification models are already in use to support the urologists in their clinical work.

View Article: PubMed Central - PubMed

Affiliation: Institute of Medical Informatics, Charité - Universitätsmedizin, Hindenburgdamm 30, 12200 Berlin, Germany. joergwichard@web.de

ABSTRACT
We investigate the performance of different classification models and their ability to recognize prostate cancer in an early stage. We build ensembles of classification models in order to increase the classification performance. We measure the performance of our models in an extensive cross-validation procedure and compare different classification models. The datasets come from clinical examinations and some of the classification models are already in use to support the urologists in their clinical work.

Show MeSH

Related in: MedlinePlus

For every partition of the cross-validation, the data is divided in a training and a test set. The performance ofeach ensemble model was assessed on validation set which was initially removed and never included in model training.
© Copyright Policy - open-access
Related In: Results  -  Collection


getmorefigures.php?uid=PMC2366047&req=5

fig2: For every partition of the cross-validation, the data is divided in a training and a test set. The performance ofeach ensemble model was assessed on validation set which was initially removed and never included in model training.

Mentions: We compared the model classes described above in a unified framework under fair conditions.Thus, we trained an ensemble of each model class consisting of 11 ensemblesmembers (11 CV-folds in the training scheme described in Section 4). Theperformance of each ensemble model was assessed on the 20% of data (validationset), which was initially removed and never included in model training (seeFigure 2). This procedure was independently repeated 20 times. This means thatall model-building processes, that is, the random removal of 20% of the data,the construction of a classification model ensemble on the remaining 80% of thedata as outlined in Section 4, and the final test on the unseen validation datawere performed each time. Finally, the mean average prediction values withrespect to the validation set were calculated and are listed in Table 2. In some cases,it is useful to apply a kind of data preprocessinglike balancing. If the distribution of the two classes differ in the sense,that one class is only represented with a small number of examples, we canbalance the data in the training set. This can improve the convergence ofseveral training algorithms and has also an impact to the classification error[35]. We applybalancing in the way that we reduce the number of samples in the one classuntil we have an balanced ratio of the class labels. The ratio of the classlabels in the validation set was never changed because it reflects the realdata distribution. Balancing was only applied to the training data. We usedthree different performance measures in order to compare the differentclassification models. Therefore, we have to define the four possible outcomesof a classification that can be formulated in a 2 × 2 confusionmatrix, as shown in Table 1. Theaccuracy,(15)Accuracy=tp+tntp+tn+fp+fn,seems to be the canonical errormeasure for almost all classification problems if the dataset is balanced.Other important measures are the specificity that quantifies how well a binaryclassification model correctly identifies the negative cases (non-PCapatients),(16)Specificity=tntn+fp,and the sensitivity, which isthe proportion of true positives of all diseased cases (PCa patients) in thepopulation,(17)Sensitivity=tptp+fn.A high sensitivity is requiredwhen early diagnosis and treatment are beneficial,which is the case in PCa.


Classification models for early detection of prostate cancer.

Wichard JD, Cammann H, Stephan C, Tolxdorff T - J. Biomed. Biotechnol. (2008)

For every partition of the cross-validation, the data is divided in a training and a test set. The performance ofeach ensemble model was assessed on validation set which was initially removed and never included in model training.
© Copyright Policy - open-access
Related In: Results  -  Collection

Show All Figures
getmorefigures.php?uid=PMC2366047&req=5

fig2: For every partition of the cross-validation, the data is divided in a training and a test set. The performance ofeach ensemble model was assessed on validation set which was initially removed and never included in model training.
Mentions: We compared the model classes described above in a unified framework under fair conditions.Thus, we trained an ensemble of each model class consisting of 11 ensemblesmembers (11 CV-folds in the training scheme described in Section 4). Theperformance of each ensemble model was assessed on the 20% of data (validationset), which was initially removed and never included in model training (seeFigure 2). This procedure was independently repeated 20 times. This means thatall model-building processes, that is, the random removal of 20% of the data,the construction of a classification model ensemble on the remaining 80% of thedata as outlined in Section 4, and the final test on the unseen validation datawere performed each time. Finally, the mean average prediction values withrespect to the validation set were calculated and are listed in Table 2. In some cases,it is useful to apply a kind of data preprocessinglike balancing. If the distribution of the two classes differ in the sense,that one class is only represented with a small number of examples, we canbalance the data in the training set. This can improve the convergence ofseveral training algorithms and has also an impact to the classification error[35]. We applybalancing in the way that we reduce the number of samples in the one classuntil we have an balanced ratio of the class labels. The ratio of the classlabels in the validation set was never changed because it reflects the realdata distribution. Balancing was only applied to the training data. We usedthree different performance measures in order to compare the differentclassification models. Therefore, we have to define the four possible outcomesof a classification that can be formulated in a 2 × 2 confusionmatrix, as shown in Table 1. Theaccuracy,(15)Accuracy=tp+tntp+tn+fp+fn,seems to be the canonical errormeasure for almost all classification problems if the dataset is balanced.Other important measures are the specificity that quantifies how well a binaryclassification model correctly identifies the negative cases (non-PCapatients),(16)Specificity=tntn+fp,and the sensitivity, which isthe proportion of true positives of all diseased cases (PCa patients) in thepopulation,(17)Sensitivity=tptp+fn.A high sensitivity is requiredwhen early diagnosis and treatment are beneficial,which is the case in PCa.

Bottom Line: We build ensembles of classification models in order to increase the classification performance.We measure the performance of our models in an extensive cross-validation procedure and compare different classification models.The datasets come from clinical examinations and some of the classification models are already in use to support the urologists in their clinical work.

View Article: PubMed Central - PubMed

Affiliation: Institute of Medical Informatics, Charité - Universitätsmedizin, Hindenburgdamm 30, 12200 Berlin, Germany. joergwichard@web.de

ABSTRACT
We investigate the performance of different classification models and their ability to recognize prostate cancer in an early stage. We build ensembles of classification models in order to increase the classification performance. We measure the performance of our models in an extensive cross-validation procedure and compare different classification models. The datasets come from clinical examinations and some of the classification models are already in use to support the urologists in their clinical work.

Show MeSH
Related in: MedlinePlus