The diagnosis of osteoporosis is eventually converted to the measurement of bone mineral density (BMD) in clinical trials. Since our previous work had proved the ability of using photoacoustic spectral analysis (PASA) to efficiently detect osteoporosis, in this contribution, we proposed a fully connected multi-layer deep neural network combined with PASA to semi-quantify BMD values corresponding to varying degrees of bone loss and to further evaluate the degree of osteoporosis. Experiments were carried out on swine femur heads, and the performance of our proposed method is satisfying for future clinical screening.