The dramatic increase in fragility fractures and the related health and economic burden rise the urge of a cutting-edge perspective to anticipate catastrophic fracture propagation in human bones. Recent studies address the issue from a multi-scale perspective, elevating the micro-scale phenomena as the key for detecting early damage occurrence. However, several limitations arise specifically for defining a quantitative framework to assess the contribution of lacunar micro-pores to fracture initiation and propagation. Moreover, the need for high resolution imaging imposes time-demanding post-processing phases. Here, we exploit synchrotron scans in combination with micro-mechanical tests, to offer a fracture mechanics-based approach for quantifying the critical stress intensification in healthy and osteoporotic trabecular human bones. This is paired with a morphological and densitometric framework for capturing lacunar network differences in presence of pathological alterations. To address the current time-consuming and computationally expensive manual/semi-automatic segmenting steps, we implement convolutional neural network to detect the initiation and propagation of micro-scale damages. The results highlight the intimate cross talks between toughening and weakening phenomena at micro-scale as a fundamental aspect for fracture prevention.