Abstract: Aiming at the geometric difference of sensitivity matrix， a clustering-based method for optimizing sensitivity matrix was proposed. Firstly， the influence of geometrical difference of sensitivity matrix on MIT image quality was analyzed. Then， based on the geometric difference， the vector of the sensitivity matrix was clustered， and the energy function was applied to assign different weights to the grouped sensitivity vector， and a clustering optimization sensitivity matrix was constructed. Finally， the optimized sensitivity matrix was used to reconstruct the MIT image with the linear back projection algorithm and the Newton-Raphson(NR) iterative algorithm. Experimental results showed that with the sensitivity matrix of clustering optimization， the mean square error of the linear back projection algorithm is reduced by more than 26%， the image correlation coefficient is increased by more than 10%， the mean square error of NR iterative algorithm is decreased by more than 5%， and the correlation coefficient is increased by more than 4%， which proves the effectiveness of the proposed method.

Key words: sensitivity matrix, geometric difference, energy function, linear back projection, Newton-Raphson(NR)