May 14, 2019


The Next Generation in Deep Learning Identification


We are very excited to announce the launch of our new, proprietary deep learning identification model, EonEndu.

What does it do?

EonEndu harnesses the power of machine learning to read radiology reports and identify patients with abnormal incidental findings (findings that may be unrelated to the reason the patient is having the scan) – specifically for Abdominal Aortic Aneurysm (AAA), Pancreatic Cysts, Pulmonary Nodules and Thyroid Nodules. Often times when incidentals are found, they are less likely to be monitored over time because hospitals do not have a streamlined way to manage them. So EonEndu not only identifies the issues, it also registers the patients to the EonDirect dashboard so they can be tracked longitudinally – which benefits both patients and hospitals.  EonEndu also uses advanced mechanisms for data extraction and data automation.


Talk Data To Me

At Eon, we specialize in bringing advanced data analysis and management techniques to the medical industry, with a focus on identifying powerful applications for machine learning and artificial intelligence.

EonEndu uses deep learning to support Natural Language Processing (NLP). Deep learning is a subfield of machine learning that is concerned with algorithms inspired by the neural networks in our brains. Unlike other machine learning models, deep learning models have networks capable of learning unsupervised (without a data scientist stepping in) from data that is unstructured or unlabeled.

NLP refers to training a model to understand natural language, or the language we humans use to communicate with one another. One common example of NLP can be found in email spam filters. Many email services use NLP text classification to “read” the content of the email and determine which category it belongs to: spam or not spam.


How EonEndu Works

EonEndu uses classic (unsupervised) NLP for a number of tasks including classifying text documents, analyzing sentiment and making recommendations. These tasks allow the model to determine things like the level of risk, the level of patient discomfort and the most common medications prescribed.

But EonEndu doesn’t stop there. The model feeds those NLP outputs into a Convolutional Neural Network (CNN) in order to retune the model with deep (supervised) learning. The outputs of those models are then vetted by leading physicians, analyzed and course corrected. This process is repeated until the results have a 90%+ positive predictive value.


Why It’s (Much) Better

This unique two-part process has resulted in the most accurate model for identifying abnormal incidental findings.

The reason for the radical difference comes back to the approach. Many techniques used to identify abnormal incidental findings rely on NLP to do a keyword search. But this approach results in much higher false positive rates. EonEndu uses NLP (unsupervised learning) initially to build models based on a dataset. However, those results are then taken and used to retrain the model using deep learning CNNs (and supervised learning). This further refines the model – resulting in a predictive value over 90%.

In short, EonEndu cuts through the hype and cost of AI by choosing the most efficient algorithm that offers the highest accuracy based on experience, experimentation and expert validation.