ADSI-MIMO: adaptive stain imputation with multi-input and multi-output learning for multiplex immunofluorescence imaging

Multiplex immunofluorescence (mIF) imaging plays a crucial role in studying multiple biomarkers and their interactions within the tumour microenvironment. However, acquiring mIF images that include all desired biomarkers presents significant challenges due to the need for specialised equipment and costly reagents, which increase technical complexity, time and expense. Stain imputation offers a promising solution by synthesising target biomarker images using generative models, thereby eliminating the need for additional staining procedures. Existing deep learning-based stain imputation methods lack flexibility in generating multiple biomarker images from various input combinations. To overcome this limitation, we propose AdSI-MIMO, a novel stain imputation framework for mIF images. Our method features a multibranch deep learning architecture capable of generating multiple outputs and incorporates an adaptive progressive masking strategy to accommodate the varying combinations of input biomarkers. This approach not only improves the quality of the generated biomarker images, but also eliminates the need to train separate models for each target biomarker. We evaluated AdSI-MIMO using two datasets, including a local dataset comprising mIF images from 257 melanoma patients and a public dataset of 55 urothelial carcinoma samples. Our method achieved substantial improvements over state-of-the-art methods, particularly in the imputation of key T-cell and activation biomarkers, such as CD8 and PD-L1. Specifically, across both public and proprietary datasets, our model achieved a 18.4% improvement in Pearson Correlation Coefficient (Pearson-r) for CD8 imputation and a 48.1% improvement for PD-L1 imputation under various input biomarker configurations on the external test set.