Pulmonary Pathology Detection: An AI-Based Approach with Convolutional Neural Networks

EasyChair Preprint 16005

Abstract

Early diagnosis of lung diseases, through the early identification of conditions, is essential to increase patient survival. This study proposes an action-research approach that combines Convolutional Neural Networks (CNNs), ResNet152, and Vision Transformers, optimized using Data Augmentation techniques. Following the ISO 25059:2023 standard, critical aspects such as data interpretability, robustness, and security are analyzed. Initial findings indicate an increase in diagnostic accuracy, exceeding 93%, thanks to advanced methods and the use of heat maps that facilitate the interpretation of results. These maps allow physicians to identify relevant areas in the analyzed images. Furthermore, they are considered essential ethical elements to ensure the privacy and security of patient information. This project aims to reduce diagnostic errors and encourage the safe adoption of artificial intelligence in hospital settings, with future steps focused on clinical validation and continuous feedback.

Keyphrases: Inteligencia Artificial, Norma ISO 25059:2023, Redes Neuronales Convolucionales (CNNs), ResNet152, Vision Transformers, data augmentation

@booklet{EasyChair:16005,
  author    = {Pablo Jose Ramirez Amador},
  title     = {Pulmonary Pathology Detection: An AI-Based Approach with Convolutional Neural Networks},
  howpublished = {EasyChair Preprint 16005},
  year      = {EasyChair, 2025}}