The inverse-geometry CT(IGCT) system uses a large-area scanned source and a 2D detector array with a smaller extent in the transverse direction. The acquired IGCT data can be significantly oversampled and the samples are not equally spaced in radial distance or angle. A previously proposed reconstruction algorithm used a modified gridding method to rebin the normalized and logged projection data into parallel projections. This approach can be suboptimal if the measured rays contributing to an output sample do not have the same signal-to-noise ratio (SNR) due to each ray having a different detected number of photons (due to different incident intensities). Reconstructed images may have better SNR if we consider the SNR of each ray in rebinning step. In this paper, we propose a new method to improve the SNR in the reconstructed image. In this method, input rays with different SNR were combined in the rebinning step by using weighted-least square fitting to produce SNR efficient output projection data. We simulated two cases : uniform, and triangular profiles of the detected number of photons across the detector array. For single slice 2D imaging, SNR improvements of 1% (uniform) and 21%(triangular) were observed. Experiments were also performed with air scan data acquired from a scanned source C-arm system (NovaRay, Inc., Palo Alto, CA). In this case, we observed SNR improvement as high as 20 %, depending on the intensity non-uniformity across the detector.