Purpose <p>Content-based medical image retrieval (CBMIR) systems have gained interest as promising tools for medical practitioners to retrieve relevant information from the exponentially growing medical imaging data available. This study was aimed to develop a CBMIR system using a cross-modality image retrieval approach based on a common vector quantization codebook for five datasets.</p> Methods <p>The proposed CBMIR system utilizes differential evolution with a novel best-neighborhood-based mutation (BNM-DE) method for common vector quantization codebook generation and optimization. By utilizing the modality of the retrieved images, the proposed system then provides automated selection of the best classifier based on DenseNet-201 for disease identification.</p> Results <p>Compared with other bioinspired optimization algorithms, such as the genetic algorithm, conventional differential evolution, particle swarm optimization, whale optimization algorithm, and grey wolf optimizer, the proposed CBMIR based on the BNM-DE method achieves a high performance. Furthermore, the proposed system achieves 100% average precision for 5 retrieved images and 98.0% average precision for 10 retrieved images.</p> Conclusion <p>The proposed CBMIR system can provide support to automate similar image retrieval and disease prediction using a bioinspired evolutionary algorithm with a novel mutation strategy, BNM-DE, for codebook optimization. The proposed system enables retrieval from multiple cross-modal datasets using an optimized common codebook. Evaluation results confirm that the proposed system with the BNM-DE method outperforms the implementation with similar methods.</p> Graphical Abstract <p></p>