In this paper, dual layer multiple-input multiple-output (2x2 MIMO) systems that use the 1024-QAM modulation are studied. We build on the layered orthogonal lattice detector (LORD), which achieves optimal maximum-likelihood (ML) performance, but which has high complexity with 1024-QAM, and argue that the low-complexity version of LORD (LC-LORD) introduces a significant performance degradation, especially with high channel correlation. We propose several approaches that outperform LC-LORD at a lower complexity. These approaches are based on improving the location of a reduced region of search within a 1024-QAM, as well was optimizing the bit log likelihood ratio (LLR) approximation. The proposed approaches were studied in the context of non-iterative and iterative detection and decoding, and significant gains were achieved in both cases.