Visualizing the pre-visual: Rice blast infection signals revealed

Disentangling pathogen infection signals in plants is critical for understanding the physiological processes that underlie the complex host-pathogen interactions and predicting impending disease outbreak. The rapid progression of rice blast lesions, caused by the filamentous fungus Magnaporthe oryzae, and its imperceptible disease-related symptoms during the asymptomatic stages render real-time detection and visualization challenging. Efforts to reveal pre-visual disease symptoms are of both broad concern and significant interest but remain challenging, as subtle disease signals are often obscured or diluted by other factors at asymptomatic stage. We introduce an imaging spectroscopy-based purification methodology that isolates the disease signals revealed by spectral unmixing on a pixel basis without considering the complex pathogen-induced physiological variations. With multi-temporal proximal hyperspectral imagery, our method captured the transition of disease lesions from asymptomatic to severely symptomatic stages, and successfully distinguished the subtle pathogen-induced signals with few false alarms as early as three days (two days after inoculation, DAI 2) before visual lesions became apparent (DAI 5). The lesion prediction results were confirmed by extensive in vivo visual inspections. Remarkably, we demonstrated that spatially aggregating the isolated disease signals improved the accuracy of pre-visual RB identification to a remarkable level up to 93 % (F1-score = 0.91), enabling unprecedented visualization of potential lesions in a narrow time window of pathogen infection. Although limitations remain regarding model validation and scalability for broader applications, this method represents a significant advancement in early disease forecasting across spectral and spatial domains, and offers new opportunities for high-throughput screening of susceptible varieties in next-generation plant resilience phenotyping.