Convolutional Neural Networks Enriched by Handcrafted Attributes (Enriched-CNN): An Innovative Approach to Pattern Recognition in Histological Images

Luiz Fernando Segato dos Santos, Leandro Neves, Alessandro Santana Martins, Guilherme Roberto, Thaína Tosta, Marcelo Zanchetta do Nascimento

2025

Abstract

This paper presents a novel method called Enriched-CNN, designed to enrich CNN models using handcrafted features extracted from multiscale and multidimensional fractal techniques. These features are incorporated directly into the loss function during model training through specific strategies. The method was applied to three important histological datasets for studying and classifying H&E-stained samples. Several CNN architectures, such as ResNet, InceptionNet, EfficientNet, and others, were tested to understand the enrichment behavior in different scenarios. The best results achieved accuracy rates ranging from 93.75% to 100% for enrichment situations involving only 3 to 5 features. This paper also provides significant insights into the conditions that most contributed to the process and allowed competitiveness compared to the specialized literature, such as the possibility of composing models with minimal or no structural changes. This unique aspect enables the method to be applied to other types of neural architectures.

Paper Citation

in Harvard Style

Santos L., Neves L., Martins A., Roberto G., Tosta T. and Zanchetta do Nascimento M. (2025). Convolutional Neural Networks Enriched by Handcrafted Attributes (Enriched-CNN): An Innovative Approach to Pattern Recognition in Histological Images. In Proceedings of the 27th International Conference on Enterprise Information Systems - Volume 1: ICEIS; ISBN 978-989-758-749-8, SciTePress, pages 467-478. DOI: 10.5220/0013277300003929

in Bibtex Style

@conference{iceis25,
author={Luiz Santos and Leandro Neves and Alessandro Martins and Guilherme Roberto and Thaína Tosta and Marcelo Zanchetta do Nascimento},
title={Convolutional Neural Networks Enriched by Handcrafted Attributes (Enriched-CNN): An Innovative Approach to Pattern Recognition in Histological Images},
booktitle={Proceedings of the 27th International Conference on Enterprise Information Systems - Volume 1: ICEIS},
year={2025},
pages={467-478},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0013277300003929},
isbn={978-989-758-749-8},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 27th International Conference on Enterprise Information Systems - Volume 1: ICEIS
TI - Convolutional Neural Networks Enriched by Handcrafted Attributes (Enriched-CNN): An Innovative Approach to Pattern Recognition in Histological Images
SN - 978-989-758-749-8
AU - Santos L.
AU - Neves L.
AU - Martins A.
AU - Roberto G.
AU - Tosta T.
AU - Zanchetta do Nascimento M.
PY - 2025
SP - 467
EP - 478
DO - 10.5220/0013277300003929
PB - SciTePress