- Missing Data May Contribute to Health Disparities in Children with Lupus
- Genome Mapping Method Could Accelerate Precision Medicine for Lupus
Lupus nephritis can be detected via blood or urine tests, but kidney biopsy is considered the most precise diagnostic approach. However, interpreting biopsy reports can be a challenge, leading to differing interpretations among multiple pathologists reading the same report.
The researchers posit that AI may help tackle these challenges.
“Given that this critical diagnostic step - which is important for planning treatment - is highly variable and imprecise, we sought out alternative approaches,” said Chandra Mohan, the Hugh Roy and Lillie Cranz Cullen Endowed Professor of Biomedical Engineering at the University of Houston, in the press release. “This funding allows us to use artificial intelligence approaches to train a ‘neural network’ to learn how to read and classify lupus nephritis biopsy slides.”
The project will help develop a computer vision pipeline to classify lupus nephritis using histopathology imaging data. In doing so, the researchers hope to create a computer-aided diagnosis system to provide clinical decision support.
“By leveraging the power of computer vision and deep learning, a branch of machine learning, we will build classifiers that rival the best renal pathologists in making a diagnosis using current criteria. This could dramatically improve patient management and long-term renal and patient outcome,” Mohan explained.
In the future, the researchers aim to develop stand-alone systems that can be leveraged as an alternative to diagnosis from a clinician.
This research is one of multiple efforts to improve lupus care through the use of sophisticated analytics technologies.
In 2021, a research team from the Medical University of South Carolina designed a machine learning (ML) approach to predict treatment response in lupus nephritis patients.
Patients are typically treated for the condition using one of two immunosuppressive drugs. These drugs are trialed over a six-month period, after which a clinician can evaluate how the patient is responding to the treatment.
Patients who do not respond to the first round of therapy are given second-line drugs, but these can be expensive, and kidney damage can occur after trialing a treatment that is ineffective.
The ML was developed to help address this by forecasting the likelihood that a patient will respond to therapy within one year using seven indicators, and the model achieved promising results.