Integrating Machine Learning and Cell Imaging to Predict Multiple Sclerosis Treatment Outcomes

Scientists in Brazil, collaborating with French institutions, have pioneered a new approach to improve the treatment of multiple sclerosis (MS) by developing a predictive tool that assesses patient response to natalizumab, a widely used MS medication. Natalizumab, a monoclonal antibody, works by blocking immune cells from migrating into the brain, thus reducing inflammation and disease progression. However, approximately 35% of patients do not respond adequately, experiencing a recurrence of symptoms within two years and suffering from side effects such as infections, headaches, fatigue, and mood disturbances.

The innovative methodology combines high-content cell imaging (HCI) with machine learning algorithms to better understand individual treatment responses. HCI employs advanced microscopy and automated image analysis to extract detailed cellular features like shape, size, and actin organization. Researchers applied natalizumab in vitro to blood samples from MS patients and analyzed over 400 morphological characteristics, narrowing down to 130 relevant traits. Machine learning models then generated millions of combinations to predict treatment outcomes.

The study achieved about 92% accuracy in the initial discovery group and 88% in a validation group, with CD8⁺ T cells identified as a significant biomarker. Patients who did not respond to therapy displayed resistant actin remodeling and more elongated, misshapen cells, indicating that their immune cells maintained a state conducive to migration, undermining the drug’s effectiveness.

This approach's results, published in Nature Communications, highlight the potential of integrating high-content imaging with machine learning to personalize MS treatment. According to lead researcher Beatriz Chaves, this strategy can improve patient quality of life by reducing unnecessary side effects, treatment delays, and costs—especially significant given natalizumab's high expense in Brazil, where it costs around BRL 10,000 per month per patient.

Future plans involve expanding the sample size, validating the model across different populations, and increasing accessibility by using simpler, more affordable imaging equipment. There is also potential to apply this methodology to other diseases, including cancer and other autoimmune disorders.

MS is a complex autoimmune neurological disease affecting the central nervous system, with symptoms ranging from muscle weakness and mobility issues to cognitive and emotional challenges. With an estimated 2.8 million affected globally and about 40,000 in Brazil, early and personalized treatment strategies are vital. The integration of machine learning and cell imaging offers a promising pathway toward more targeted, effective, and less toxic therapies.