Photoacoustic tomography (PAT) is a valuable tool in characterizing ovarian lesions for accurate diagnosis. However, limited view problem degrades the quality of PAT reconstruction severely, especially for transvaginal transducer which only partially encloses the target. To address this issue, we compensated limited view information loss by co-registered PAT and US machine learning method. The simulation and phantom results showed that the details of the target were recovered by proposed method, compared with delay-and-sum reconstruction method.