Classification of H&E Images via CNN Models with XAI Approaches, DeepDream Representations and Multiple Classifiers

• Main Author: Neves, Leandro Alves [UNESP]
• Publication Date: 2023
• Other Authors: Martinez, João Manuel Cardoso [UNESP], da Costa Longo, Leonardo H. [UNESP], Roberto, Guilherme Freire, Tosta, Thaína Aparecida Azevedo, de Faria, Paulo Rogério, Loyola, Adriano Mota, Cardoso, Sérgio Vitorino, Silva, Adriano Barbosa, do Nascimento, Marcelo Zanchetta, Rozendo, Guilherme Botazzo [UNESP]
• Format: Conference object
• Language: eng
• Source: Repositório Institucional da UNESP
• Download full: http://dx.doi.org/10.5220/0011839400003467; http://hdl.handle.net/11449/248918
• topic: Classification; DeepDream Representations; Grad-CAM; Histological Images; LIME

The study of diseases via histological images with machine learning techniques has provided important advances for diagnostic support systems. In this project, a study was developed to classify patterns in histological images, based on the association of convolutional neural networks, explainable artificial intelligence techniques, DeepDream representations and multiple classifiers. The images under investigation were representatives of breast cancer, colorectal cancer, liver tissue, and oral dysplasia. The most relevant features were associated by applying the Relief algorithm. The classifiers used were Rotation Forest, Multilayer Perceptron, Logistic, Random Forest, Decorate, IBk, K*, and SVM. The main results were areas under the ROC curve ranging from 0.994 to 1, achieved with a maximum of 100 features. The collected information allows for expanding the use of consolidated techniques in the area of classification and pattern recognition, in addition to supporting future applications in computer-aided diagnosis.