Aidan Yuen-Oye : Deep Learning for Brain Metastasis Radiosurgery скачать в хорошем качестве

Aidan Yuen-Oye : Deep Learning for Brain Metastasis Radiosurgery

Masters proposal presentation: Aidan Yuen-Oye: Implementation of Deep Learning Models for Delineation and Assessment of Treatment Effect Following Radiosurgery for Brain Metastasis Aims This project aims to develop a deep learning model to streamline the SRS planning workflow and improve detection, classification and therefore treatment for brain metastases. To achieve this overall goal, we specifically aim to: 1. Develop a deep learning model to automatically segment brain metastases on MRI scans. 2. Improve the accuracy of early lesion volume estimation compared to current clinician rough early approximations. 3. Assess the utility of automated contours for peer review and detection of missed lesions. 4. Develop and validate a classification algorithm using radiomic features to differentiate between tumour progression and radiation necrosis. 5. Streamline the stereotactic radiosurgery planning workflow by integrating automated segmentation and classification tools. Significance The application of DLMs in Radiation Oncology has the potential to substantially enhance workflow efficiency and clinical decision-making. In practice, an automatic segmentation model could be applied to patient MRI scans acquired for SRS treatment planning. These DLM generated contours would provide a foundation for clinicians to refine, greatly reducing the time required for manual contouring. By standardising this initial step, inter-observer variability would be minimised, and treatment planning practices would become more consistent. Moreover, such a system could act as a built-in quality assurance mechanism, functioning as a second observer for peer-to-peer review to decrease the likelihood of missed lesions. Importantly, the ability to generate early volumetric estimates would also allow oncologists to make more informed treatment fractionation decisions at an earlier stage, ultimately streamlining scheduling and facilitating more timely treatment delivery. Beyond segmentation, a radiomics-based classification model capable of distinguishing tumour progression from RN would provide additional clinical value. Radiomic feature extraction and machine learning may offer superior accuracy in lesion characterisation compared to the interpretation of a single radiation oncologist. Improved diagnostic confidence could reduce dependence on MDT review and invasive neurosurgical biopsies, while enabling earlier and more definitive clinical decision-making. This would translate directly into improved patient outcomes by ensuring that progressive disease is treated promptly and that patients with RN are spared unnecessary interventions. For more information visit our: Website: https://www.uwamedicalphysics.org Weblog: http://www.uwamedicalphysics.com Facebook:   / groups   Instagram:   / uwa_medical_physics   LinkedIn:   / medical-physics-uwa-11979b379   Twitter: @MedicalUwa YouTube Channel:    / @medicalphysicsuwa   =========================