Oladosu Oladimeji
Oladosu Oyebisi Oladimeji is a Doctoral Researcher in the Computer Vision and Machine Learning Research Group led by Dr Saritha Unnikrishnan at Atlantic Technological University, Ireland. As a member of the ATU-Unnikrishnan-Lab, his research focuses on artificial intelligence for breast cancer screening and diagnosis using digital mammography, digital breast tomosynthesis, contrast-enhanced mammography and ultrasound. His work and the group’s published project repositories are available at: https://github.com/ATU-Unnikrishnan-Lab
Oladosu holds an M.Sc. (Hons) in Computer Science from the University of Ibadan, Nigeria, and previously served as an Assistant Lecturer and Researcher at KolaDaisi University, Ibadan. He is the author or co-author of over 30 peer-reviewed journal articles and conference papers spanning breast cancer imaging, brain tumour classification, explainable AI, and digital mental health, published in journals.
His research has been supported by the MOCHAS Postgraduate Research Training Programme, a British Society of Breast Radiology (BSBR) research bursary (2025), and a World Technology Universities Network exchange award for collaborative work with the University of Bradford on AI for nanotherapeutics. He was part of the winning team at the National AI Challenge 2024 (TechIreland, supported by Google, OpenAI, and Enterprise Ireland).
Beyond research, Oladosu teaches at ATU, reviews for journals including Computers in Biology and Medicine and Biomedical Signal Processing and is active in STEM outreach and mentoring.
-------------------------------------------------------------------------------------------------
Title: Adaptive Cross-View Gating for Robust, Calibrated Classification of BI-RADS 4 Breast Lesions
BI-RADS 4 lesions carry a malignancy probability spanning 2–95%, yet biopsy is recommended across this entire range. Roughly two-thirds of biopsied BI-RADS 4 lesions prove benign, exposing patients to unnecessary physical, psychological, and financial burden. Radiologists routinely interpret craniocaudal (CC) and mediolateral oblique (MLO) views together, but most deep learning systems process each view independently, discarding the cross-view context clinicians rely on and rarely account for views being missing or incomplete, as occurs in real practice.
This study introduces an adaptive cross-view gating mechanism that allows each mammographic view to selectively recalibrate the other, mirroring how radiologists use one projection to contextualise findings on the second. This was evaluated against simpler fusion strategies (concatenation, addition, maximum pooling) across two independent datasets, CDD-CESM (contrast-enhanced mammography) and CBIS-DDSM (digitised screen-film mammography) using patient-level splits and repeated 5-fold cross-validation.
Gating-based fusion consistently outperformed simple fusion on both datasets. An ablation study confirmed the gating mechanism itself, rather than added complexity, drives this gain. Clinically, gating strategies produced better-calibrated probability estimates essential for biopsy decision support and degraded more gracefully than simple fusion when a view was missing at inference, a common real-world scenario.
Cross-view gating offers a trustworthy, clinically-motivated fusion paradigm for multi-view mammographic AI, with direct relevance to reducing unnecessary biopsies in BI-RADS 4 decision-making. This work is supported by the BSBR Research Bursary and is under review at the International Journal of Imaging Systems and Technology.
Managed and payment processed by:
KC Jones conference&events Ltd
Telephone: 01332 227773
Email: bsbr@kc-jones.co.uk