382MO Enhancing phase I clinical trial selection using artificial intelligence: Evaluation of a large language model algorithm in a dedicated drug development unit

Abstract

Methods: Here, we present such a pipeline that can work with 3D surface scans from consumer-level devices, such as iPhones with LiDAR sensors.Our approach combines AI-based breast detection in 2D images with a back-projection method that maps segmentations onto 3D meshes, enabling accurate volume calculation.The system is currently being trained on more than 8,000 annotated images from the REQUITE dataset (www.requite.eu)to improve generalisability across diverse imaging conditions.Results: Validation on diverse 3D meshes demonstrates robust performance, opening up prospects for a clinical trial to evaluate accuracy on real scans.Importantly, the pipeline is fully automated, transparent in its intermediate steps, and designed to be adaptable to additional datasets, supporting continuous refinement. Conclusions:This pipeline is a first step toward objective and accessible breast volumetry.By removing reliance on specialised hardware, costly software, or "blackbox" algorithms, it lowers adoption barriers and empowers both patients and clinicians with bias-free measurements.Beyond being part of cosmetic evaluation, such open-source tools could guide reconstruction, enable longitudinal monitoring, and become a backbone for a model capable of predicting future appearance.

Ähnliche Arbeiten