As a new molecular imaging technology, Cherenkov-excited luminescence scanned imaging (CELSI) has merits of high spatial resolution and large imaging depth, therefore showing a potential for monitoring the physiological changes of tumors during radiotherapy. In our previous work, we developed a tomographic technique for CELSI based on Tikhonov method, which is problematic to reconstruct accurate fluorescent targets with position depth larger than 3 cm or with low contrast. To overcome this problem, we develop a sparse reconstruction method for tomographic CELSI based on approximate message passing. To demonstrate the merits of the proposed algorithm, we compare it with traditional Tikhonov regularization and three sparse based reconstruction algorithms. Our results show that the proposed method can achieve best performance in terms of mean-square error and contrast noise ratio.