Diagnosis of Brain Tumors Using a Convolutional Neural Network

DelaCruz-VdV P.; Cadenillas-Rivera D.; Vega-Huerta H.; Cancho-Rodriguez E.; Bulnes M.E.P.; Maquen-Niño G.L.E.; Pantoja-Collantes J. · Smart Innovation, Systems and Technologies (2023) · DOI: 10.1007/978-981-99-5414-8_6

Resumen

The purpose of this research is to support the diagnosis of brain tumors in adult patients. To obtain greater accuracy in the detection of the diagnosis, the following input variables were considered: type; class; grade; location and size of the brain tumor. To achieve the proposed objective, a convolutional neural network composed of three levels of layers was implemented: (a) the input layer, which captures the images and applies a filter or kernel that returns a map of the characteristics of the original image, thus reducing the size of the parameters; (b) the reduction or pooling layer whose function is to reduce the spatial dimensions of the input volume; and (c) a connected classifier layer: this classifier layer will have as many neurons as the number of classes to be predicted. A total of 2870 images were used to support the implementation and training of the convolutional neural network model and thus identify whether the patient's brain was healthy or had a brain tumor, for this diagnosis an image of a real case of a patient with a brain tumor was taken, and the result was the identification of the type of tumor “pituitary”, i.e., the convolutional neural network effectively diagnosed the brain tumor. The results obtained indicate that the trained model was effective and efficient. To validate the effectiveness of the algorithm, the confusion matrix was used, which yielded an accuracy of 0.9059. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2023.

Instituciones: Universidad Nacional Mayor de San Marcos; Universidad Ricardo Palma; Universidad Nacional Pedro Ruiz Gallo