Visual cortex speckle imaging for shape recognition.

This study introduces a non‑invasive approach for neurovisual classification of geometric shapes by capturing and decoding laser‑speckle patterns reflected from the human striate cortex. Using a fast digital camera and deep neural networks (DNN), we demonstrate that each visual stimulus - rectangle, triangle, mixed shapes, or blank screen, arouses a detectably distinct speckle pattern. Our optimized DNN classifier achieved near perfect recall (98%) for rectangles and high recall (91%) for triangles in single‑shape trials and sustained robust performance (82% recall) when multiple shapes appeared simultaneously. Circular stimuli produced subtler and less reliable speckle dynamics and were not classified with consistent accuracy. By leveraging low‑cost optics and scalable AI processing, this technique paves the way for real‑time, portable monitoring of visual cortex activity, offering transformative potential for cognitive neuroscience, brain-machine interfaces, and clinical assessment of visual processing. Future work will expand stimulus complexity, optimize model architectures, and explore multimodal neurophotonic applications.