Întotdeauna devotat.

Displaying items by tag: neural networks

patrocinadores1-300x72 1

One of the oldest - and yet actual - medical challenges is the prediction of the most probable outcome of a disease in a given patient, based on the clinical and pathology information and a therapy scenario.

A model based on Artificial Neural Networks (ANN) was designed and programmed. The input data (over 120 parameters) include histology type and grade of tumours, morphometry and fractal dimension of microscopic images of lesions, clinical and para-clinical data, quality of life, and treatment. The key output is life expectation for dogs and cats with cancer, but many input parameters can be turned into unknowns and the network asked to provide an estimate.

The ANN was tested on a smaller set of 27 criteria and 39 cases of cancer in dogs to develop appropriate architecture and learning strategies. Robustness and predictive performance were confirmed. As previously reported, we also found overfitting/overtraining to be the most serious pitfall that needs to be addressed. The complete model is growing and learning.

ANN are one very promising way to respond to the growing interest for Evidence Based Medicine methods applied in veterinary practice. ANN provide a lean approach for integrating in the current diagnostic and prognostic procedures some new or still ‘exotic’ pathology information, like fractal dimensions of histology images.

Journal of Comparative Pathology, Volume 148, Issue 1, January 2013, Page 69

Full poster presentation can be viewed here.                                            High resolution pdf here.

Authors: dr. Liviu Gaita, Prof. dr. Manuella Militaru


Published in Oncologie

Descoperiți Cronicile ORTOVET Excelsior. Scrise, fotografiate și filmate zi de zi...