Classification of H&E images exploring ensemble learning with two-stage feature selection

• Main Author: Tenguam, Jaqueline Junko [UNESP]
• Publication Date: 2022
• Other Authors: Da Costa Longo, Leonardo Henrique [UNESP], Silva, Adriano Barbosa, De Faria, Paulo Rogerio, Do Nascimento, Marcelo Zanchetta, Neves, Leandro Alves [UNESP]
• Format: Conference object
• Language: eng
• Source: Repositório Institucional da UNESP
• Download full: http://dx.doi.org/10.1109/IWSSIP55020.2022.9854418; http://hdl.handle.net/11449/241593
• topic: ensemble learning; feature selection; histological images; ranking with metaheuristics

In this work, an investigation based on ensemble learning is presented for the recognition of patterns in histological tissues stained with Hematoxylin and Eosin, representative of breast cancer, colorectal cancer, liver tissues and oral dysplasia. The strategy considered compositions with multiple descriptors, such as deep learned and handcrafted, and multiple classifiers. The deep learned descriptors were calculated by exploring different architectures of convolutional neural networks. The handcrafted descriptors were representative of the multidimensional and multiscale fractal categories, Haralick and local binary pattern. The main combinations were obtained through two-stage feature selection (ranking with wrapper selection) and classified via an ensemble composed of five classifiers. The accuracy rates were values between 93.10% and 100%, with some highlights involving the main combinations of approaches.