Abstract 3684: Integrating multiparametric MRI with spatial transcriptomics to identify “Radio-Spatial Genomic” features of prostate cancer using artificial intelligence

Abstract Introduction: Risk stratification remains a key challenge in prostate cancer (PCa) management involves risk stratification, and identification of the subgroup of patients at highest risk of progressing from localised to metastatic disease is critical. Multiparametric MRI (mpMRI) is key in the PCa diagnostic pathway. By integrating clinical parameters, mpMRI radiomics and spatial transcriptomics (ST), this novel “Radio-Spatial Genomics” platform offers an exciting opportunity to identify mpMRI radiomic features associated with important biological aspects of PCa linked to an aggressive disease phenotype. Methods: Multi-regional spatial transcriptomics (Visium 10x Genomics) was performed on archived formalin-fixed paraffin-embedded prostatectomy sections from patients recruited to a local trial (ISRCTN10046036). Axial sections were sequenced using ST (8 per patient) from 2 patients with Gleason 4+4 PCa and preoperative mpMRI available was used for this study. Anatomical landmarks on mpMRI were segmented by a radiologist. Using a proportional size algorithm and a convolutional neural network (ProsRegNet), T2-axial MRI slices were aligned and registered to histopathology sections. An application, SpatialStitcher, was developed on Python 3.7.0 to digitally stitch separate ST sections for image registration. Results: Using the prostatic capsule and urethra as landmarks, histopathology sections from 2 patients were co-registered to corresponding T2-axial MRI slices. In total, a median of 114670 whole transcriptome sequenced barcoded ST spots were co-registered to 30424 pixels on MRI per patient. A median DICE correlation score of 0.942, 0.738 and 0.756 was achieved for capsule, tumour and BPH nodules respectively. AMACR (marker for PCa) expression inversely correlated with T2 MRI intensity-based radiomic features (r = -0.763), consistent with the tumour being hypointense. Differential gene expression analysis between hyperintense peri-tumoural and hypointense tumour regions revealed enrichment for genes involved in mucosal immune response. Conclusion: In this study, we report preliminary results of mapping MRI with ST using machine learning to identify genotypic changes based on radiomics. This novel “Radio-Spatial Genomics” model may allow the detection of clinically relevant genotypic features from diagnostic prostate mpMRI imaging. Citation Format: Thineskrishna Anbarasan, Sandy Figiel, Sophia M. Abusamra, Wencheng Yin, Nithesh Ranasinha, James T. Grist, Dan J. Woodcock, Richard J. Bryant, Ruth McPherson, Freddie C. Hamdy, Bartlomiej Papiez, Ian G. Mills, Alastair D. Lamb. Integrating multiparametric MRI with spatial transcriptomics to identify “Radio-Spatial Genomic” features of prostate cancer using artificial intelligence [abstract]. In: Proceedings of the American Association for Cancer Research Annual Meeting 2025; Part 1 (Regular Abstracts); 2025 Apr 25-30; Chicago, IL. Philadelphia (PA): AACR; Cancer Res 2025;85(8_Suppl_1):Abstract nr 3684.