A GA-Based CNN Model for Brain Tumor Classification

ÖZDEM K., Ozkaya Ç., ATAY Y., ÇELTİKÇİ E., BÖRCEK A. Ö., Demirezen U., ...More

7th International Conference on Computer Science and Engineering, UBMK 2022, Diyarbakır, Turkey, 14 - 16 September 2022, pp.418-423 identifier

  • Publication Type: Conference Paper / Full Text
  • Doi Number: 10.1109/ubmk55850.2022.9919461
  • City: Diyarbakır
  • Country: Turkey
  • Page Numbers: pp.418-423
  • Keywords: Cancer, classification, CNN, genetic algorithm, hyperparameter, machine learning, MRI, tumor
  • Ankara Yıldırım Beyazıt University Affiliated: No


Detection and classification of tumor types generally cover problem-specific algorithm developments. The problems of detecting tumors with the analysis of standard brain images obtained with different medical imaging tools and frequently used in the literature are always desired, developed, and discussed. This study focuses on identifying tumors, extracting different characteristics, and associating them with cancer types. The standard approach of convolutional neural networks (CNN) was used primarily for the identification of tumors. Then, the genetic algorithm (GA) approach was designed and used for hyperparameter optimization in CNN to increase the performance in all datasets. Thus, a CNN+GA hybrid model was proposed and analyzed with different tests. In this process, the results were examined in detail and the standard CNN algorithm and some machine learning methods suggested in the literature were analyzed comparatively. In addition, the data set called Gazi Brains 2020 Dataset, which was obtained within the scope of the Turkish Brain Project, is also used to test the proposed system. Here, literature reviews of the previous studies in which different machine/deep learning approaches are used together with optimization algorithms are presented. The different comparison scores obtained according to the experimental studies were presented in the tables and the outputs were evaluated in terms of significance. The results have shown that the proposed hybrid models are successful in achieving better accuracies not only with different datasets available in the literature but also DL/ML models trained with Gazi Brain 2020 Dataset. It should be concluded that the proposed method might be also used for other deep/machine learning models and applications.