Ruusuvuori lab
At the 30th anniversary Medical Image Understanding and Analysis conference in Dublin, Pekka presented a new paper by Hesam Hakimnejad, Sonja Koivukoski, Leena Latonen and Pekka: “Residual Shifting-Based Diffusion Models for Virtual H&E Staining”. The method offers a more computationally efficient way to generate virtual H&E stains from unstained tissue.
Kaisa Liimatainen, Leena Latonen and Pekka published “SparStVR, exploring sparse 3D histology data in virtual reality” in Communications Engineering. The VR application lets you interactively explore 3D histology data with multimodal heatmap support.
Umair Khan, Jouni Härkönen, Marjukka Friman, Hesam Hakimnejad, Leena Latonen, Teijo Kuopio and Pekka published a multicenter benchmarking study on staining normalization in histopathology in Scientific Reports. The paper compares traditional and generative AI normalization methods and tests how robust foundation models are to staining variation.
Hyder Abbas, Sonja Koivukoski, Hesam Hakimnejad, Zsombor Malatinszki, Tuomas Mäkilä, Leena Latonen and Pekka won the Best Paper Award at the 6th IEEE Conference on ICT Solutions for eHealth in Vilamoura, Portugal, for “Transfer Learning Based on Generative Adversarial Network for Tissue-Specific Virtual Staining”. The group also presented a second paper at the conference, “StainFlow: Virtual Staining of Label-Free Tissue via Flow-Based Generative Models”, by Hesam Hakimnejad, Sonja Koivukoski, Leena Latonen and Pekka.
Pekka co-authored a new AI model for predicting DNA mismatch repair (dMMR) status in colorectal cancer from histopathology, published in Computer Methods and Programs in Biomedicine. Led by Liisa Petäinen and Jussi Väyrynen, the model uses non-tumor and low-magnification tissue regions to cut analysis time from days to minutes, and was validated across cohorts from Finland and the US.
Pekka and Masi contributed to a new bioRxiv preprint on how hypoxia and acidosis shape myeloid cell responses in glioblastoma, led by Kirsi Rautajoki’s team. The study combines cyclic immunohistochemistry, single-cell RNA sequencing, spatial transcriptomics and DNA methylation profiling across 136 glioblastomas, showing that infiltrating macrophages adapt to survive acidic hypoxic niches while microglia are depleted there and lose their homeostatic identity.