 Home 5 Clinical Diagnostics Insider 5 4 Open-Source Diagnostic Software Projects to Explore

4 Open-Source Diagnostic Software Projects to Explore

by Holden Galusha | Jun 30, 2023 | Clinical Diagnostics Insider, Diagnostic Testing and Emerging Technologies, Special Focus-dtet

Free open-source software can be leveraged across diagnostic processes—how can your organization benefit from it?

Diagnostic testing labs often face two hurdles when adopting new software: cost and rigidity. The highly specialized software used in diagnostic processes can be cost-prohibitive. Other times, the software may not meet process requirements adequately, making it even harder to justify the cost. Open-source software is an alternative that addresses both challenges.

Open-source software has publicly available source code, which means it can be downloaded and modified by users to better suit their needs at no cost.

While free and open-source software (FOSS) often comes with downsides such as a lack of technical support options or unpolished user experiences, in the right circumstances it can be an excellent avenue for budget-conscious labs to explore. Here are four open-source projects developed for diagnostic processes such as medical imaging, data extraction, image archival, and insulin dose regulation.

Project MONAI

Project Medical Open Network for AI, or MONAI, is a suite of development frameworks for building medical imaging applications powered by artificial intelligence (AI). Conceived by Kings College London in partnership with NVIDIA—the computing giant in graphics technology, AI, and more—the purpose of MONAI is to “accelerate the pace of innovation and clinical translation.”¹

Who uses MONAI?

MONAI was created for healthcare researchers who develop deep learning models for use in medical imaging applications. MONAI offers solutions across the domains of imaging, deep learning, and software deployment with three frameworks: Core, Label, and Deploy.

MONAI Core

Core is the “flagship library of Project MONAI,” offering functions for domain-specific training of medical imaging AI models such as:

image preprocessing functions (or “image transforms,” as they’re referred to in the machine learning lexicon) to ready them for analysis by an AI model,

3D segmentation algorithms, which are used to identify elements of interest within 3D medical scans for analysis, and

a prebuilt machine learning (ML) framework that addresses common drawbacks with other medical image segmentation models, such as long search times.²

With MONAI Core, medical imaging researchers will be well-equipped to develop novel AI-driven imaging applications.

MONAI Label

MONAI Label is an intelligent image labeling system that lessens the burden on researchers to manually annotate datasets for use in training ML models. Radiologists mark up scans in DICOM images regularly, but these annotations aren’t necessarily compatible with the requirements of training an ML model. According to the website, “By utilizing user interactions, MONAI Label trains an AI model for a specific task and continuously learns and updates that model as it receives additional annotated images.”¹

MONAI Label supports radiology and pathology images and can be integrated into DICOM viewers for both.

MONAI Deploy

The team behind MONAI intends for the Deploy framework to become the “de-facto standard” for developing and implementing medical AI applications within clinical settings. Deploy offers a modular architecture for building a custom deployment pipeline that will scale to your lab’s needs.

medspaCy

It’s common for clinical researchers to use natural language processing (NLP), a form of AI that can process human language, to extract data stored in electronic medical record (EMR) systems. However, adapting existing NLP offerings for unique clinical research workflows can be difficult and time-consuming.³ This is where solutions like medspaCy come in.

Developed by researchers from the University of Utah, medspaCy is an implementation of the popular NLP framework spaCy optimized to process clinical text. According to the project’s GitHub page, “The medspaCy package brings together a number of other packages, each of which implements specific functionality for common clinical text processing specific to the clinical domain…” Built using Python, a highly flexible programming language undergoing a sort of AI renaissance, spaCy “provides a framework for modular plug-and-play construction of custom NLP pipelines.”⁴

Who uses medspaCy?

Clinical researchers seeking to leverage clinical NLP (cNLP) in their workflows could benefit from medspaCy and similar solutions. By forking the spaCy code library into a variant specialized for processing clinical text, the creators of medspaCy offer clinical researchers a new avenue for easily incorporating NLP into their processes while avoiding the overhead associated with legacy cNLP solutions built using other programming languages.

Dicoogle

According to the website, Dicoogle is an open-source picture archiving and communications system (PACS) platform developed by software company BMD Software and research group UA.PT Bioinformatics. It has a modular, plugin-based architecture along with a software development kit to enable complete customization by the user.⁵

Dicoogle is also scalable. According to the project website, it has been tested with over 25 million indexed DICOM objects and is optimized for big data paradigms.⁵

It should be noted that the medical image viewer built into Dicoogle is not a professional viewer and lacks the functionality required for screening and other diagnostic activities. However, as it is open-source, this limitation can be remedied with a custom solution.

Who uses Dicoogle?

Clinical labs on a tight budget or who want to customize their PACS to function in a certain way can find uses for Dicoogle. Because it is open source, it is free and easily modifiable. Research groups, universities, and medical imaging companies have all used Dicoogle for various purposes.

Tidepool

Tidepool is a non-profit organization developing an open-source data platform to “make diabetes data more accessible, actionable, and meaningful.” Tidepool’s data collection tool, Uploader, allows people to upload data from their diabetes devices to the group’s web application, Tidepool Data Platform. Data Platform then visualizes the data in interactive dashboards.⁶

Patterns can be displayed in hourly, daily, and weekly windows, assisting in identifying insulin usage patterns and more. All of this data serves to help the patient have more effective, insightful conversations about treatment with their healthcare provider. Tidepool’s mobile app also assists with data tracking and note-taking.⁷

Additionally, Tidepool recently unveiled the Tidepool Loop, the “first fully interoperable automated insulin dosing app” cleared by the FDA. With the Loop app, those with diabetes will be able to manage their insulin levels more easily—reducing the risk of overdosing or missing doses. While still unavailable to consumers, the Loop app will be compatible with certain Apple Watches.⁷

Who uses Tidepool?

Those with diabetes, as well as their healthcare providers, can benefit from Tidepool. The Tidepool Loop’s automated dosing will help patients manage insulin doses, while the Tidepool Data Platform can help both patients and clinicians make informed decisions about treatment.

FOSS can be a boon for clinicians, researchers, and others involved with diagnostic processes. Cost benefits aside, the flexibility afforded by open codebases allows for limitless customization to tailor a solution to an organization’s unique workflow.