A child suffering from specific language impairment (SLI) or developmental dysphasia (DD) has delayed speech and disordered language development with no apparent reason. Unlike existing methods that use only acoustic features for screening, a new approach is proposed where the texture present in the pathological speech signal is captured by local binary patterns (LBP) from the joint time-frequency representation. The LBP extracts the latent information present in the temporal domain which is generally not captured by the standard audio features. Experiments were performed on a database having 4214 utterances from 44 healthy and 54 SLI subjects. Experimental results indicate that with the proposed system the classification accuracy rate of the system has been improved up to 13.1% if the LBP features are fused along with the traditional features with a highest average accuracy rate of 97.36% when a 5-fold cross validation is done.