Existing categorical grading methods in medicine fail to capture subtle variations in disease spectrum which are crucial for closely monitoring its progression, response to treatment, and developing new guidelines to enable personalized and precision medicine. We developed a truly continuous, non-categorical, numerical score, called the Edema Score (ES), with a precision of multiple decimal places, for grading macular edema in retinal optical coherence tomography (OCT) images. This score was determined by obtaining the directional derivatives using activations extracted from the final convolutional layer of a deep learning retinal OCT model and the specialist defined edema concepts. Significant differences in ES were observed between different groups of edema severity (p < 0.001). It captured complex nonlinear relationships from edema features in an image, that correlated well with the disease severity, response to treatment, the prognostically significant internal pattern of edema, and its invisible spread into the surrounding regions. The scalability of this deep radiomics approach to more biomarkers in Ophthalmic imaging and in other domains of medicine, can serve as a bridge ushering us into an era of precision medicine.