Robust Deep Learning-based 3-D Scene Reconstruction and Segmentation for Improved Diagnosis of Medical Ailments in Surgical Videos