AI maps immune patterns in TB-infected mouse lungs

Tuberculosis (TB) is a complex infection that produces organized immune structures in the lung called granulomas. These structures can differ widely depending on whether an infection is acute, chronic, or asymptomatic, making it hard for researchers to pin down which immune patterns are linked to disease progression or control. To tackle that challenge, a team led by corresponding author Usama Sajjad applied a computational imaging approach to see immune cells and their relationships inside granulomas. Working in Diversity Outbred Mice, the researchers combined high-content imaging with automated analysis so they could look beyond single-cell counts and study the spatial layout of immune cells. The study used Deep Learning-Based Spatial Immunoprofiling on multiplex immunofluorescence images to detect and quantify cell types and their positions within lung lesions. By doing so the team aimed to distinguish disease states not merely by presence of cells but by how those cells are organized relative to one another inside the granuloma.

The core methods relied on machine learning applied to detailed fluorescence images of lung tissue. Using Deep Learning-Based Spatial Immunoprofiling, the group processed multiplex immunofluorescence images to identify and map immune cells across granulomas. From those maps they measured cellularity and proximity relationships among cell types. The results showed clear differences by disease state: acute TB featured dysfunctional responses with low cellularity, while chronic TB had the highest numbers of immune cells. Importantly, asymptomatic infection produced a distinct pattern — granulomas in this state had increased T cell density, increased numbers of peribronchiolar B cells, and T cells positioned closer to macrophages compared to both acute and chronic TB. The study also established an automated pipeline to extract and analyze data from multiplexed fluorescence images, making the spatial measurements reproducible and scalable.

The work helps set a practical foundation for studying how granuloma architecture varies by disease state. By emphasizing spatial relationships — for example, how close T cells are to macrophages or where B cells collect around bronchioles — the approach moves beyond simple counts of immune cells to a richer picture of tissue organization. The automated pipeline the team developed can reduce manual effort and bias in image analysis and allow researchers to compare many lesions and animals with consistent criteria. While carried out in Diversity Outbred Mice, the findings underscore that different TB states leave distinctive spatial signatures in lung granulomas. These signatures could be used by scientists to better link immune organization with infection outcomes and to design experiments that test how altering cell positions affects disease progression.