BEGIN:VCALENDAR VERSION:2.0 PRODID:-//Talks.cam//talks.cam.ac.uk// X-WR-CALNAME:Talks.cam BEGIN:VEVENT SUMMARY:Cambridge MedAI Seminar - September 2025 - Dr Tammar Truzman and M r Artsiom Hramyka DTSTART:20250930T104500Z DTEND:20250930T120000Z UID:TALK236686@talks.cam.ac.uk CONTACT:Hannah Clayton DESCRIPTION:Sign up on Eventbrite: https://medai_september2025.eventbrite. co.uk\n\nJoin us for the *Cambridge AI in Medicine Seminar Series*\, hoste d by the *Cancer Research UK Cambridge Centre* and the *Department of Radi ology at Addenbrooke’s*. This series brings together leading experts to explore cutting-edge AI applications in healthcare—from disease diagnosi s to drug discovery. It’s a unique opportunity for researchers\, practit ioners\, and students to stay at the forefront of AI innovations and engag e in discussions shaping the future of AI in healthcare.\n\nThis month’s seminar will be held on *Tuesday 30 September 2025\, 12-1pm at the Jeffre y Cheah Biomedical Centre (Main Lecture Theatre)\, University of Cambridge * and *streamed online via Zoom*. A light lunch from Aromi will be served from 11:45. The event will feature the following talks:\n\n*_Automated Les ion Segmentation of Stroke MRI Using nnU-Net: A Comprehensive External Val idation_ - Dr Tammar Truzman\, Postdoctoral Fellow\, MRC Cognition and Bra in Sciences Unit\, University of Cambridge*\n\nDr Tammar Truzman is a Post doctoral Fellow at the MRC Cognition and Brain Sciences Unit\, University of Cambridge\, working with Prof. Matt Lambon Ralph and Dr. Ajay Halai. He r research focuses on language assessment and recovery in people with apha sia\, combining neuroimaging\, language rehabilitation\, and computational modeling. She is also a licensed speech-language pathologist with experti se in language therapy and clinical translation.\n\n*Abstract:* Accurate l esion segmentation is a critical step in stroke neuroimaging\, both for ad vancing theoretical understanding of brain–behavior relationships and fo r enabling clinical applications. Deep learning methods have recently show n promise\, but external validation across diverse datasets remains limite d. In this talk\, I will present a comprehensive evaluation of nnU-Net for stroke lesion segmentation across multiple acute and chronic datasets. I will discuss factors influencing model performance and generalization\, in cluding imaging modality\, dataset size and quality and lesion volume. The results highlight both the potential and the current limitations of autom ated segmentation tools for translational use in stroke and aphasia resear ch.\n\n*_Deep Learning-Based Follicle Growth Prediction using a Transforme r Architecture_ - Artsiom Hramyka\, Postdoctoral Fellow\, University of Ca mbridge*\n\nArtsiom is a Postdoctoral Researcher in Computer Science and M edicine at the University of Cambridge\, where his work involves applying artificial intelligence and simulation modelling to solve complex healthca re problems. This research builds upon his doctoral work at the University of St Andrews\, where he is completing his PhD thesis on the application of novel analytical frameworks and AI in healthcare. Currently\, his prima ry focus is on the early detection of cancer as part of the CRUK Internati onal Alliance for Cancer Early Detection (ACED). In this role\, he develop s and calibrates multistate models that simulate the natural history of ma lignant cancers to evaluate and optimise screening strategies. His researc h also extends to other areas of medicine\, including active collaboration s where he applies machine learning to enhance fertility treatments with I mperial College London and to analyse treatment data in paediatric oncolog y with the Charlotte Maxeke Johannesburg Academic Hospital.\n\n*Abstract:* Traditional methods for predicting ovarian follicle growth rely on the cl inically unfeasible assumption of tracking individual follicles between ul trasound scans. This research introduces a novel approach that overcomes t his limitation by predicting the entire follicle size distribution. We dev eloped a decoder-only\, GPT-like Transformer architecture to autoregressiv ely forecast future follicle profiles from sequential scan data. Model per formance was evaluated using distribution-level metrics\, including Earth Mover's Distance (EMD) and Chi-Square distance\, across three clinically r elevant scenarios simulating different data availability. Systematic hyper parameter optimisation resulted in a performance increase\, a 10.2% improv ement in EMD for short-term predictions. A key finding is the robust perfo rmance of the model when using only a single initial scan\, demonstrating its potential utility in cases with missed appointments and highlighting t he importance of training-inference consistency. This work represents the first application of a Transformer architecture for distribution-level fol licle prediction\, offering a more realistic tool for clinical decision su pport in assisted reproductive technology.\n\n\nThis is a hybrid event so you can also join via Zoom:\n\nhttps://zoom.us/j/99050467573?pwd=UE5OdFdTS FdZeUtIcU1DbXpmdlNGZz09\n\nMeeting ID: 990 5046 7573 and Passcode: 617729\ n\nWe look forward to your participation! If you are interested in getting involved and presenting your work\, please email Ines Machado at im549@ca m.ac.uk\n\nFor more information about this seminar series\, see: https://w ww.integratedcancermedicine.org/research/cambridge-medai-seminar-series/\n LOCATION:Jeffrey Cheah Biomedical Centre (Main Lecture Theatre)\, Universi ty of Cambridge END:VEVENT END:VCALENDAR