.jpg)
Artificial intelligence-assisted diagnosis in radiology has seen remarkable advances, especially in the automatic classification of thoracic diseases from chest X-rays. However, the presence of multiple pathologies with overlapping visual features poses a significant technical challenge for classification systems. To address this complexity, recent research has proposed hierarchical multi-label classification approaches that leverage taxonomic relationships between different conditions. Two novel techniques, loss-based and logit-based, integrate clinical hierarchy knowledge directly into the optimization process or adjust predicted probabilities according to the parent class. Results on massive datasets such as CheXpert and NIH show notable improvements in precision, AUC, and F1, outperforming reference methods by 10% to 24%. This type of innovation not only improves performance but also provides more interpretable results for clinical decision-making.
Implementing these systems in real-world environments requires robust and customized technological solutions. At Q2BSTUDIO, as a company specialized in custom software, we develop artificial intelligence platforms for businesses that integrate hierarchical classification models into hospital workflows. Our services range from creating custom applications to orchestrating infrastructures on AWS and Azure cloud services, ensuring scalability and regulatory compliance. Additionally, we combine AI agents with business intelligence tools like Power BI to visualize results and apply cybersecurity measures to protect sensitive data. This comprehensive approach allows advances in multi-label classification to translate into operational and secure solutions, accelerating diagnosis and improving patient care.
