Bell Eapen MD, PhD.

Bringing Digital health & Gen AI research to life!

Named Entity Recognition using LLMs: a cTakes alternative?

TLDR: The targeted distillation method described may be useful for creating an LLM-based cTakes alternative for Named Entity Recognition. However, the recipe is not available yet. 

Named Entity Recognition using LLMs: a cTakes alternative?

Image credit: Wikimedia

Named Entity Recognition is essential in clinical documents because it enhances patient safety, supports efficient healthcare workflows, aids in research and analytics, and ensures compliance with regulations. It enables healthcare organizations to harness the valuable information contained in clinical documents for improved patient care and outcomes. 

Though Large Language Models (LLMs) can perform Named Entity Recognition (NER), the capability can be improved by fine-tuning, where you provide the model with input text that contains named entities and their associated labels. The model learns to recognize these entities and classify them into predefined categories. However, as described before fine-tuning Large Language Models (LLMs) is challenging due to the need for substantial, high-quality labelled data, the risk of overfitting on limited datasets, complex hyperparameter tuning, the requirement for computational resources, domain adaptation difficulties, ethical considerations, the interpretability of results, and the necessity of defining appropriate evaluation metrics. 

Targeted distillation of Large Language Models (LLMs) is a process where a smaller model is trained to mimic the behaviour of a larger, pre-trained LLM but only for specific tasks or domains. It distills the essential knowledge of the LLM, making it more efficient and suitable for particular applications, reducing computational demands.  

This paper described targeted distillation with mission-focused instruction tuning to train student models that can excel in a broad application class. The authors present a general recipe for such targeted distillation from LLMs and demonstrate that for open-domain NER. Their recipe may be useful for creating efficient distilled models that can perform NER on clinical documents, a potential alternative to cTakes. Though the authors have open-sourced their generic UniversalNER model, they haven’t released the distillation recipe code yet. 

REF: Zhou, W., Zhang, S., Gu, Y., Chen, M., & Poon, H. (2023). UniversalNER: Targeted Distillation from Large Language Models for Open Named Entity Recognition. ArXiv. /abs/2308.03279 

Published by Bell Eapen on | Permalink

Kedro for multimodal machine learning in healthcare 

Healthcare data is heterogenous with several types of data like reports, tabular data, and images. Combining multiple modalities of data into a single model can be challenging due to several reasons. One challenge is that the diverse types of data may have different structures, formats, and scales which can make it difficult to integrate them into a single model. Additionally, some modalities of data may be missing or incomplete, which can make it difficult to train a model effectively. Another challenge is that different modalities of data may require different types of pre-processing and feature extraction techniques, which can further complicate the integration process. Furthermore, the lack of large-scale, annotated datasets that have multiple modalities of data can also be a challenge. Despite these challenges, advances in deep learning, multi-task learning and transfer learning are making it possible to develop models that can effectively combine multiple modalities of data and achieve reliable performance. 

Pipelines Kedro for multimodal machine learning

Kedro for multimodal machine learning

Kedro is an open-source Python framework that helps data scientists and engineers organize their code, increase productivity and collaboration, and make it easier to deploy their models to production. It is built on top of popular libraries such as Pandas, TensorFlow and PySpark, and follows best practices from software engineering, such as modularity and code reusability. Kedro supplies a standardized structure for organizing code, handling data and configuration, and running experiments. It also includes built-in support for version control, logging, and testing, making it easy to implement reproducible and maintainable pipelines. Additionally, Kedro allows to easily deploy the pipeline on cloud platforms like AWS, GCP or Azure. This makes it a powerful tool for creating robust and scalable data science and data engineering pipelines. 

I have built a few kedro packages that can make multi-modal machine learning easy in healthcare. The packages supply prebuilt pipelines for preprocessing images, tabular and text data and build fusion models that can be trained on multi-modal data for easy deployment. The text preprocessing package currently supports BERT and CNN-text models. There is also a template that you can copy to build your own pipelines making use of the preprocessing pipelines that I have built. Any number and combination of data types are supported. Additionally, like any other kedro pipeline, these can be deployed on kubeflow and VertexAI. Do comment below if you find these tools useful in your research. 

Template for multi-modal machine learning in healthcare using Kedro. Combine reports, tabular data and images using various fusion methods.
https://github.com/dermatologist/kedro-multimodal
3 forks.
19 stars.
0 open issues.

Recent commits:

Dark Mode

kedro-multimodal (this link opens in a new window) by dermatologist (this link opens in a new window)

Template for multi-modal machine learning in healthcare using Kedro. Combine reports, tabular data and image using various fusion methods.

1 Subscriber 3 Watchers 0 Forks Check out this repository on GitHub.com (this link opens in a new window)

Published by Bell Eapen on | Permalink

Six things data scientists in healthcare should know

Healthcare, like most other fields, is eager to get on the data science bandwagon. Data scientists can make a huge difference in the way big data is utilized for clinical decision-making. However, there are paradigmatic differences in the way data scientists from quantitative fields view the world, compared to their clinical counterparts. This is especially true in the emerging fields of machine learning and artificial intelligence. This may lead to considerable inefficiencies. As a person trained in both fields, here is my take on this.

Data scientists
Credit: Dasaptaerwin, CC0, via Wikimedia Commons

Data scientists should focus on the problem and not the solutions

Data scientists are excited about the latest GPT or BERT. Data scientists tend to refine the model a bit more using 10 more GPUs! In the process, they tend to solve problems that do not exist. From my experience practicing medicine in extremely resource-poor areas, simple solutions are valued more than BERT running on Kubernetes! This is true in the developed world as well, and many teams may have fundamental data needs that need to be tackled first.

Explanation comes before prediction

Emerging machine learning methods prioritize prediction accuracy compromising on explainability in the process. Clinicians, in most cases, cannot use nor trust a model that arrives at a conclusion without showing how it reached there. Hence, in the clinical domain, a simple logistic regression model may be more acceptable than a deep learning neural network. Parsimony is the key and a bit of feature selection to ensure parsimony will be appreciated always.

You need to know the clinical terminologies

A basic understanding of the clinical terminologies and terminology systems such as SNOMED and ICD is vital. It helps in understanding the clinical community better. Any healthcare analytics to consider variations in terminologies and adopt a standard system for consistency. Any tool that data scientists build for the clinical community should have support for terminology systems.

Biostatistics is more pervasive than you think

Most healthcare professionals are trained in biostatistics. Hence, the thinking leans towards population, sampling, randomization, blindings and showing a ‘statistically significant’ difference. Moving towards machine learning needs a paradigmatic shift. It may be useful to have a discussion on this at the outset.

Classes are of unequal importance

In healthcare, finding one class (e.g. cancer) is more important than the other class (e.g. no cancer). One class may need active intervention to save lives. Hence, sensitivity and specificity are of vital importance than accuracy!

Life is precious!

In healthcare, there is no room for error. Some decisions may have disastrous consequences while few others may save lives. As a data scientist in the healthcare domain, you should be cognizant of the fact that healthcare data is different from banking/airline data.

Published by Bell Eapen on | Permalink

How to deploy an h2o ai model using OpenFaaS on Digitalocean in 2 minutes

H2O is an open-source, distributed and scalable machine learning platform written in JAVA. H2O supports many of the statistical & machine learning algorithms, including gradient boosted machines, generalized linear models, deep learning and more.  OpenFaaS® (Functions as a Service) is a framework for building Serverless functions easily with Docker. Read my previous post to learn more about OpenFaaS and DO. 

H2O AI model deployment

H2O has a module aptly named sparkling water that allows users to combine the machine learning algorithms of H2O with the capabilities of Spark. Integrating these two open-source environments provides a seamless experience for users who want to make a query using Spark SQL, feed the results into H2O to build a model and make predictions, and then use the results again in Spark. For any given problem, better interoperability between tools provides a better experience.

H2O Driverless AI is a commercial package for automatic machine learning that automates some of the most difficult data science and machine learning workflows such as feature engineering, model validation, model tuning, model selection, and model deployment. H2O also has a popular open-source module called AutoML that automates the process of training a large selection of candidate models. H2O’s AutoML can be used for automating the machine learning workflow, which includes automatic training and tuning of many models within a user-specified time-limit. AutoML makes hyperparameter tuning accessible to everyone.

H2O allows you to convert the models to either a Plain Old Java Object (POJO) or a Model Object or an Optimized (MOJO) that can be easily embeddable in any Java environment. The only compilation and runtime dependency for a generated model is the h2o-genmodel.jar file produced as the build output of these packages. You can read more about deploying h2o models here.

I have created an OpenFaaS template for deploying the exported MOJO file using a base java container and the dependencies defined in the gradle build file. Using the OpenFaaS CLI (How to Install) pull my template as below:

mkdir watersplash
cd watersplash

faas-cli template pull https://github.com/dermatologist/java-ext --prefix your-docker-uname

faas-cli new --lang java-h2o watersplash

Copy the exported MOJO zip file to the root folder along with build.gradle and settings.gradle. Make appropriate changes to handle.java as per the needs of the model, as explained here. Add http://digitaloceanIP:8080 to watersplash.yml

 provider:
  	name: openfaas
  	gateway: http://digitaloceanIP:8080

and finally:

 faas-cli up -f watersplash.yml

That’s it! Congratulations! Your model is up and running! Access it at http://digitaloceanIP:8080/function/watersplash

If you get stuck at any stage, give me a shout below. 

Published by Bell Eapen on | Permalink

Machine Learning on Diabetic Retinopathy Images

Artificial intelligence (AI) and Machine Learning (ML) are having a profound impact on the way medicine is being practiced. AI/ML algorithms and techniques fit imaging applications easily and can help with automation. Radiology is the specialty that has benefitted the most from the AI/ML revolution. Melanoma detection in Dermatology is another obvious winner.

Image credit: pixabay.com

Many of the machine learning algorithms are reasonably well known. The real challenge is to get the infrastructure to crunch massive amounts of data, getting the ideal dataset for a problem, optimizing the model for performance and deploying the model for use. If you are relatively new to ML, Kaggle is a useful resource for you to start.

I will briefly introduce Kaggle for those who have not used it before. Kaggle is a platform for posting datasets that you have collected. They also provide ‘kernels’ or computational resources (typically Jupyter Notebooks) for collaborative analysis. The datasets can be made private or public under a variety of license options. Organizations post competitions and reward teams that solve them. Solutions are typically posted as predictions on a test dataset or share the kernel code

I recently noticed a good competition on Kaggle that the eHealth community may find interesting. Aravind Eye Hospital in India has posted a dataset consisting of fundoscopic images of diabetic retinopathy with varying degrees of severity. The dataset consists of thousands of images collected in rural areas by the technicians of Aravind hospital from the rural areas of India. The challenge is to develop a model that can predict the severity of diabetic retinopathy from the fundoscopic image. Further, the successful solutions will be shared with other Ophthalmologists through the 4th Asia Pacific Tele-Ophthalmology Society (APTOS) Symposium.

The competition page is available here: https://www.kaggle.com/c/aptos2019-blindness-detection
Let me know if anybody wants to team up!

Published by Bell Eapen on | Permalink

Serverless on FHIR: Management guidelines for the semi-technical clinician!

Serverless is the new kid on the block with services such as AWS Lambda, Google Cloud Functions or Microsoft Azure Functions. Essentially it lets users deploy a function (Function As A Service or FaaS) on the cloud with very little effort. Requirements such as security, privacy, scaling, and availability are taken care of by the framework itself. As healthcare slowly yet steadily progress towards machine learning and AI, serverless is sure to make a significant impact on Health IT. Here I will explain serverless (and some related technologies) for the semi-technical clinicians and put forward some architectural best practices for using serverless in healthcare with FHIR as the data interchange format.

artificial intelligence
Serverless on FHIR

Let us say, your analyst creates a neural network model based on a few million patient records that can predict the risk for MI from BP, blood sugar, and exercise. Let us call this model r = f(bp, bs, e). The model is so good that you want to use it on a regular basis on your patients and better still, you want to share it with your colleagues. So you contact your IT team to make this happen.

This is what your IT guys currently do: First, they create a web application that can take bp, bs and e as inputs using a standard interface such as REST and return r. Next, they rent a virtual machine (VM) from a cloud provider (such as DigitalOcean). Then they convert this application into a container (docker) and deploy it in the VM. You now can use this as an application from your browser (chrome) or your EMR (such as OpenMRS or OSCAR) can directly access this function. You can share it with your colleagues and they can access it in their browsers and you are happy. The VM can support up to 3 users at a time.

In a couple of months, your algorithm becomes so popular that at any one time hundreds of users try to access it and your poor VM crashes most of the time or your users have to wait forever. So you call your IT guys again for a solution. They make 100 copies of your container, but your hospital is reluctant to give you the additional funding required.

Your smart resident notices that your application is being used only in the morning hours and in the night all the 100 containers are virtually sleeping. This is not a good use of the funding dollars. You contact your IT guys again, and they set up Kubernetes for orchestrating the containers according to usage. So, what is Serverless? Serverless is a framework that makes all these so easy that you may not even need your IT guys to do this. (Well, maybe that is an exaggeration)

My personal favourite serverless toolset (if you care) is Kubernetes + Knative + riff. I don’t try to explain what the last two are or how to use them. They are so new that they keep changing every day. In essence, your IT team can complete all the above tasks with few commands typed on the command line on the cloud provider of your choice. The application (function rather) can even scale to zero! (You don’t pay anything when nobody uses it and add more containers as users increase, scaling down in the night as in your case).

Best Practices

What are the best practices when you design such useful cloud-based ‘functions’ for healthcare that can be shared by multiple users and organizations? Well, here are my two cents!

First, you need a standard for data exchange. As JSON is the data format for most APIs, FHIR wins hands down here.

Next, APIs need a mechanism to expose their capabilities and properties to the world. For example, r = f(bp, bs, e) needs to tell everyone what it accepts (bp, bs, e) and what it returns (at the bare minimum). FHIR has a resource specifically for this that has been (not so creatively) named as an Endpoint. So, a function endpoint should return a FHIR Endpoint resource with information about itself if there is no payload.

What should the payload be? Payload should be a FHIR Bundle that has all the FHIR Resources that the function needs (bp, bs and e as FHIR Observations in your case). The bundle should also include a FHIR Subscription resource that points to the receiving system (maybe your EMR) for the response ( r ).

So, what next?

Take the phone and call your IT team. Tell them to take
Kubernetes + Knative + riff for a spin! I might do the same and if I do, I will share it here. And last but not the least, click on the blue buttons below! 🙂

Published by Bell Eapen on | Permalink

Hephestus: Health data warehousing tool for public health and clinical research

Health data warehousing is becoming an important requirement for deriving knowledge from the vast amount of health data that healthcare organizations collect. A data warehouse is vital for collaborative and predictive analytics. The first step in designing a data warehouse is to decide on a suitable data model. This is followed by the extract-transform-load (ETL) process that converts source data to the new data model amenable for analytics.

The OHDSI – OMOP Common Data Model is one such data model that allows for the systematic analysis of disparate observational databases and EMRs. The data from diverse systems needs to be extracted, transformed and loaded on to a CDM database. Once a database has been converted to the OMOP CDM, evidence can be generated using standardized analytics tools that are already available.

Each data source requires customized ETL tools for this conversion from the source data to CDM. The OHDSI ecosystem has made some tools available for helping the ETL process such as the White Rabbit and the Rabbit In a Hat. However, health data warehousing process is still challenging because of the variability of source databases in terms of structure and implementations.

Hephestus is an open-source python tool for this ETL process organized into modules to allow code reuse between various ETL tools for open-source EMR systems and data sources. Hephestus uses SqlAlchemy for database connection and automapping tables to classes and bonobo for managing ETL. The ultimate aim is to develop a tool that can translate the report from the OHDSI tools into an ETL script with minimal intervention. This is a good python starter project for eHealth geeks.

Anyone anywhere in the world can build their own environment that can store patient-level observational health data, convert their data to OHDSI’s open community data standards (including the OMOP Common Data Model), run open-source analytics using the OHDSI toolkit, and collaborate in OHDSI research studies that advance our shared mission toward reliable evidence generation. Join the journey! here

Disclaimer: Hephestus is just my experiment and is not a part of the official OHDSI toolset.

[github-clone username=”dermatologist” repository=”hephaestus”]

Published by Bell Eapen on | Permalink

LesionMapper: Pictographic lesion encoder for Dermatology

An electronic medical record example

An electronic medical record example (Photo credit: Wikipedia)

Grading systems and novel methods of symptom coding is becoming more and more important with the growth of telehealth and electronic health records. It is probable that in dermatology too, a significant number of consultations will move online soon.

Visual Analogue Scale (VAS) is a commonly used tool for measuring subjective sensations such as itching. There is evidence showing that visual analogue scales have superior metrical characteristics than discrete scales, thus a wider range of statistical methods can be applied to the measurements.

Couple of months back, I attended a thesis defense in McMaster in which an innovative web based tool called Pain-QuILTTM for visual self-report of pain was presented. The technique of iconography – pictorial material relating to or illustrating a subject – was employed to represent pain using a flash based web-interface. Pain-QuILTTM tracks quality, intensity, location and temporal characteristics of the pain. Quality is represented by different icons, intensity is represented by a visual analogue scale of 1 -10, location by the position of icon on the body image and temporal characteristics by the time stamp. The clinical feasibility of Pain-QuILTTM has been successfully validated and published (1).
Pain-QuILTTM is a property of McMaster University and is subject to McMaster University’s terms of use. It can be accessed here. 

The iconographic symptom encoding could be applied easily to dermatology as well. Dermatology lesions are primarily visual and dermatological diagnosis to a great extend is based on the type, distribution, intensity and temporal characteristics of the skin lesions. However the representation may be challenging because of the diverse nature of lesions.

Recently I came across fabric.js a javascript library for image manipulation based on HTML5 canvas. Fabric.js was much more versatile and powerful than I expected. I could prototype  LesionMapperTM (that is what I want to call it), in less than 24 hours. The type of lesions are symbolized by representative clinical pictures instead of icons, intensity is represented by the opacity/translucency of the image and the location and distribution by the position and size of the lesion respectively, on the body image. The images can be dragged, enlarged or rotated. The icing on the cake is the ability of fabric.js to rasterize the image into a JSON that can be stored easily in a database.

Update: 14- June – 2016

LesionMapper is available as an OpenMRS module and OSCAR eForm. OpenMRS module is opensource and can be downloaded here. The github repository is available here. If you need the OSCAR eForm version, please contact me.

 

Ref:
1. Lalloo, Chitra et al. “Pain-QuILT: Clinical Feasibility of a Web-Based Visual Pain Assessment Tool in Adults With Chronic Pain.” Journal of medical Internet research 16.5 (2014). [JMIR]

Published by Bell Eapen on | Permalink