Improved Medical Classification using Semantic Analysis and Data Augumentation
Hospitals and clinics perform a large number of screening tests for early illness detection and improve patient’s chances of a full recovery. However, they struggle to prioritize patients effectively and to advance them to treatment. Artificial Intelligence comes a paramount tool to assist medical staff with automatized pathology detection. The abdomen is a critical anatomic region, connecting all regions of the body and carrying many organs with distinct characteristics.
This makes it very susceptible to symptoms, becoming a bottleneck in the prioritization process. This study presents a methodology to improve a multi-class classifier by increasing the samples in the input dataset with linguistic analysis. We search keywords by pattern and generate its associated distribution that is necessary to achieve a minimum of 92% recall. The classifier is based on the Transformer architecture with Contextual Word Embedding augmentation. Our results show a 96% recall score, boosting the detection of afflicted patients.
Walter Silva Martins-Filho
Senior Data Scientist at Neuralmed
Walter Martins-Filho, MSc., is a senior data scientist with 4+ years experience in applied NLP and machine learning algorithms to medical reports and records, insurance profiles and . He is working at Neuralmed, a health tech company . He graduated in Astronomy specialized in Computational Astronomy by the Federal University of Rio de Janeiro, UFRJ. And He have a master degree in Astronomy by the National Observatory of Brazil. He formerly worked in the Lunar and Planetary Laboratory of the University of Arizona as visiting researcher, in the UFRJ as a professor, and in the Getulio Vargas Foundation, an economic think tank in Brazil, as data scientist.
