Academy

Democratizing Healthcare: How We Built Latin America's First AI Medical Diagnosis System in Spanish

While sipping mate in our Buenos Aires office back in 2020, we embarked on an ambitious journey: creating Latin America's first AI-powered medical diagnosis system in Spanish. Today, I want to share this story of innovation, social impact, and how we're working to make healthcare more accessible for the 483 million Spanish speakers worldwide.

Standing on the Shoulders of Giants

At ÜMA Health, we believe in the power of open knowledge and collaboration. While many companies were developing English-language AI diagnosis systems, Spanish speakers - including millions in the United States - were left behind with rule-based systems that showed poor accuracy.

We saw this gap as an opportunity to create something revolutionary. Just like tango emerged from the fusion of different cultural influences in the ports of Buenos Aires, we combined state-of-the-art deep learning with our understanding of local healthcare needs.

The Technical Tango

Our approach was comprehensive and novel. We analyzed 759,000 anonymized general practice records to create a sequential model with four main components:

  1. A Symptoms Translator: Converting colloquial Spanish expressions into medical terminology
  2. A Symptoms Extractor: Identifying and categorizing medical symptoms
  3. A Text Generator: Creating coherent medical narratives
  4. A Diagnosis Classifier: Determining the most likely diagnosis

Using a Hierarchical Attention Network model (based on https://www.cs.cmu.edu/~./hovy/papers/16HLT-hierarchical-attention-networks.pdf), we achieved remarkable results:

  • 82% accuracy for top diagnosis
  • 93% accuracy when considering the top three diagnoses

These numbers aren't just statistics - they represent real people getting faster, more accurate medical attention.

From Research to Reality

The beautiful thing about technology is its ability to transform theoretical research into practical solutions. We didn't want this to remain just an academic paper - we launched it as a fully functional product at uma-health.com/autonomous.

This isn't just another SaaS product - it's a tool that's helping democratize healthcare access across Latin America and Spanish-speaking communities worldwide. Just as the public library opened doors to knowledge, we're opening doors to healthcare accessibility.

Looking Forward

This research, presented at Machine Learning for Healthcare 2020, marks just the beginning. We're continuing to improve the system, expand its capabilities, and work towards our vision of universal healthcare access.

As we say in Argentina, "el camino se hace al andar" (the path is made by walking). We're proud to be walking this path, combining technological innovation with social impact, and we invite others to join us in this journey.

Remember: technology is most beautiful when it serves humanity. That's what we strive for at ÜMA Health, one line of code at a time.

Abstract here: https://static1.squarespace.com/static/59d5ac1780bd5ef9c396eda6/t/5f2440d51877ee516da5f770/1596211414528/19CameraReadySubmissionClinical_Abstract_MLH_2020.pdf