Case study
01 · The question
Challenge
Single bacteriophages appear as crowded, variable fluorescence signals. Reliable counting must remain robust across imaging conditions while retaining a meaningful relationship to the underlying molecular process.
02 · The work
Approach
I developed U-Net-based density prediction workflows, controlled synthetic data, automated HPC training pipelines, and validation strategies grounded in Poisson statistics and Langmuir binding models.
03 · The value
Impact
The work links pixel-level inference with estimates of antigen–antibody affinity, creating a path from computer vision output to interpretable biophysical measurement.