Gated Spatial Hidden Markov Models for Classification in Histological Samples, With Application to Muscle...

Quantitative histology often requires assigning each site or cell to one of two classes from continuous stain intensities. In practice, large-scale illumination and contrast fluctuations bias thresholds and obscure true biological differences. Standard correction methods preprocess the image to remove these artifacts, but do not propagate uncertainty from the correction into subsequent classification. We propose a spatial hidden Markov model that performs probabilistic, illumination-adjusted classification directly on raw stain intensities. A deterministic, data-driven "gate" rescales each site's effective decision threshold according to local brightness, while an autologistic prior encourages spatial coherence. Bayesian inference yields posterior class probabilities and parameter summaries of class separation and spatial dependence. Applications to muscle histology demonstrate that the gated SHMM yields physiologically plausible classifications that are robust to uneven illumination while preserving fine-scale biological structure.