High dimensionality in data introduces challenges in computational efficiency, interpretability, and model robustness. Linear dimensionality-reduction methods capture global variance but often miss complex nonlinear patterns. Nonlinear projection methods are more effective at preserving local relationships but are limited because most rely on Euclidean distance, which performs poorly for some biomedical data. We modified the nonlinear t-distributed Stochastic Neighbor Embedding method to use the Jensen-Shannon divergence for probabilistic neighbor estimation. This information-theoretic measure is bounded, symmetric, and robust. We applied this method to three biomedical imaging datasets: mass spectrometry imaging of skin and breast tissue, and mass cytometry imaging of triple-negative breast cancer. We hypothesized that the proposed method would improve the separation of labeled tissue regions in the reduced space. We compared our method with two conventional dimensionality-reduction methods. Our results showed that the proposed method produced more compact and visually distinct clusters. Quantitative evaluation generally confirmed a statistically significant improvement in separability and classification performance. Our results suggested that using Jensen-Shannon divergence in dimensionality reduction enhances both visualization and analysis of high-dimensional biomedical imaging data.