Using deep learning image segmentation to prepare lung masks for CT scans of COVID-19 patients

The Challenge

We had to train our algorithm from scratch and prepare a segmentation model to identify the lungs in 3D computerized tomography (CT) scans for healthy, diseased, and consolidated lung tissue better than off-the-shelf models. These other models work slice-by-slice on 2D images, which gives the lung very jagged edges and features in adjacent slices do not help the present layer.

We chose state-of-the-art 3D model structures and a novel model, using both labelled and unlabelled data to develop a performance model for mild to severe respiratory patients.

Procogia’s Approach

Hypothesis testing: our approach was to develop the models by posing hypotheses and repeatedly testing them at different points in development (e.g., whether domain-inspired data augmentation would improve model performance)
Iterative approach: we developed several iterations of models using data that was partially annotated using previous iterations and corrected by hand
Domain knowledge: we liaised with the client to inform them of our data augmentation strategy and guide the criteria for model acceptance
Define milestones: we defined several important milestones which guided the model development and performance, including: over 95% average dice score in all cases, over 95% average dice on severe COVID cases, submit abstract based on work to a conference, and host model internally.

The Results

Our model performs with >95% average dice over all test cases, compared to benchmark models, for example, the state-of-the-art work by Hofmanninger, which achieved dice scores of around 90% on our data
Our model can be used to volumetrically distinguish inhalation and exhalation CT scans with >95% F1 score. This allowed other research teams to sort their data automatically, saving days or weeks of tedious manual work
Our paper was accepted at an international biomedical conference to discuss the benefits of the data augmentation approach we used. This gave us a chance to showcase our expertise and data findings
Our model is hosted internally, providing lung segmentation to the organization, expediting the use of this model by internal company research teams to accelerate their research.

Services Used

Data Science

We use open source technology to leverage the full potential of your data. Predictive and prescriptive results are actioned using AI and Machine Learning (ML).

Machine Learning

Drawing on statistical methods to enable improvement with experience, ProCogia’s machine learning algorithms predict outcomes and automate processes.

Data Engineering

ProCogia is partnered with the leading cloud providers, enabling our agnostic approach to focus on delivering tailored game-changing solutions for our clients.

Medical Imaging

Whereas off-the-shelf medical imaging models typically use 2D images, ProCogia’s model utilizes higher-quality image segmentation in 3D CT scans to seamlessly distinguish between healthy, diseased and consolidated lung tissue.