Diffuse Optical Imaging (DOI) techniques are an ever growing field of research as they are noninvasive, compact, cost-effective and can furnish functional information about human tissues. Among others, they include techniques such as Tomography, which solves an inverse reconstruction problem in a tissue volume, and Mapping which only seeks to find values on a tissue surface. Limitations in reliability and resolution, due to the ill-posedness of the underlying inverse problems, have hindered the clinical uptake of this medical imaging modality. Multimodal imaging and Deep Learning present themselves as two promising solutions to further research in DOI. In relation to the first idea, we implement and assess here a set of methods for SOLUS, a combined Ultrasound (US) and Diffuse Optical Tomography (DOT) probe for breast cancer diagnosis. An ad hoc morphological prior is extracted from US B-mode images and utilised for the regularisation of the inverse problem in DOT. Combination of the latter in reconstruction with a linearised forward model for DOT is assessed on specifically designed dual phantoms. The same reconstruction approach with the incorporation of a spectral model has been assessed on meat phantoms for reconstruction of functional properties. A simulation study with realistic digital phantoms is presented for an assessment of a non-linear model in reconstruction for the quantification of optical properties of breast lesions. A set of machine learning tools is presented for diagnosis breast lesions based on the reconstructed optical properties. A preliminary clinical study with the SOLUS probe is presented. Finally, a specifically designed deep learning architecture for diffusion is applied to mapping on the brain cortex or Diffuse Optical Cortical Mapping (DOCM). An assessment of its performances is presented on simulated and experimental data.