Discretization is an important step introduced in the field of Knowledge Discovery in Databases to better represent the knowledge domain and increase the learning speed and performance of Data Reduction and Data Mining algorithms. However, no studies evaluated the benefits of introducing a discretization step into a more complex system. In this study we seek to evaluate how the ChiMerge discretization method could improve the performance of a CAD system for the automatic detection of prostate cancer (PCa) based on multi-parametric Magnetic Resonance (mp-MR) imaging. 16 semiquantitative and quantitative features were extracted from malignant and normal region of interest in 56 patients, who underwent mp-MR exam before prostatectomy. By using ChiMerge on a training set, we computed different cut points for each feature to transform the continuous attributes into discrete variables. Both the continuous and the discretized 16-dimensional vector generated for all voxels have been separately fed into the SVM classifier used by the CAD system and the performances were compared. Moreover, a feature selection (FS) method based on the correlation between parameters was applied to both the continuous and the discrete features, and the performances of the CAD system when using the resulting subset of features have been evaluated. Results showed that the CAD system obtained the best performance when it uses all the discretized parameters. Besides, FS applied on the discretized parameters did not affect the results obtained with all the discretized parameters (p=0.565), thus making the use of the FS method feasible to reduce dimensionality. Finally, our results showed that the discretization greatly improves the results of patients having a starting area under the ROC curve <0.75, that represents a critical situation for a CAD system. In conclusion, preliminary results show that discretization can effectively and substantially increase the performance of a CAD system.
ChiMerge Discretization Method: Impact on a Computer Aided Diagnosis System for Prostate Cancer in MRI / Rosati, Samanta; Giannini, Valentina; Mazzetti, S.; Russo, F.; Regge, D.; Balestra, Gabriella. - ELETTRONICO. - (2015), pp. 297-301. (Intervento presentato al convegno MeMeA 2015 tenutosi a Torino nel may 7-9, 2015).
ChiMerge Discretization Method: Impact on a Computer Aided Diagnosis System for Prostate Cancer in MRI
ROSATI, SAMANTA;GIANNINI, VALENTINA;BALESTRA, Gabriella
2015
Abstract
Discretization is an important step introduced in the field of Knowledge Discovery in Databases to better represent the knowledge domain and increase the learning speed and performance of Data Reduction and Data Mining algorithms. However, no studies evaluated the benefits of introducing a discretization step into a more complex system. In this study we seek to evaluate how the ChiMerge discretization method could improve the performance of a CAD system for the automatic detection of prostate cancer (PCa) based on multi-parametric Magnetic Resonance (mp-MR) imaging. 16 semiquantitative and quantitative features were extracted from malignant and normal region of interest in 56 patients, who underwent mp-MR exam before prostatectomy. By using ChiMerge on a training set, we computed different cut points for each feature to transform the continuous attributes into discrete variables. Both the continuous and the discretized 16-dimensional vector generated for all voxels have been separately fed into the SVM classifier used by the CAD system and the performances were compared. Moreover, a feature selection (FS) method based on the correlation between parameters was applied to both the continuous and the discrete features, and the performances of the CAD system when using the resulting subset of features have been evaluated. Results showed that the CAD system obtained the best performance when it uses all the discretized parameters. Besides, FS applied on the discretized parameters did not affect the results obtained with all the discretized parameters (p=0.565), thus making the use of the FS method feasible to reduce dimensionality. Finally, our results showed that the discretization greatly improves the results of patients having a starting area under the ROC curve <0.75, that represents a critical situation for a CAD system. In conclusion, preliminary results show that discretization can effectively and substantially increase the performance of a CAD system.Pubblicazioni consigliate
I documenti in IRIS sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/11583/2606199
Attenzione
Attenzione! I dati visualizzati non sono stati sottoposti a validazione da parte dell'ateneo