Optimization of Fourier Ptychographic microscope using phase images for malaria detection

Fourier ptychographic microscopy is a recent imaging technique that overcomes the limitations of conventional optics. The images it produces are particularly fine and super-resolved. They are also very rich, since they are bimodal (intensity and phase images) compared with conventional microscopy. FPM therefore holds great promise for a whole range of medical applications. In this work, the potential of this microscopy is explored by considering the biological application of automatic diagnosis of malaria on a stained blood smear. We report that an appreciable improvement in the classification of parasitized red blood cells is obtained when intensity and phase images are jointly exploited in a deep convolutionnal neural network, compared to that obtained with intensity images alone. We also show that such joint exploitation considerably relaxes the constraints relative to the choice of microscope objective. In particular, an objective lens with a numerical aperture as low as 0.2 can be used with little degradation in classification performance. The performances obtained are close to those obtained with a conventional resolution microscope equiped with a 0.9 numerical aperture objective. This can be highly desirable for the realization of rapid diagnostic system, which requires access to large fields of view.