Workflow in 2019 ESMO poster
The core of the technology is to provide end-to-end and generalized solutions for predicting the prognosis of metastatic brain tumors as rapid brain metastasis and subsequent medical procedures are critical. Human factors such as physician experience, habits, and fatigue have also added many possible errors to the process of brain metastasis. The personalized tumor labeling system will create an automated and standardized brain tumor labeling process, and provide a personalized elastic space for brain tumor contouring. The deep learning method is used to construct a recurrence prediction from the image factor and structured clinical data of the MRI image. The model can assist physicians in the choice of treatment options, and provides a Domain Learning module for Transfer Learning to provide self-learning of machine learning programs, allowing the client to optimize the predictive program based on patient characteristics.
The system integrates (1) deep learning network (3D-Vnet) as the structure of the personalized brain tumor automatic labeling system (2) deep network self-encoder as the base image and clinical biometric brain tumor recurrence prediction model ( 3) Improve the generalization effect of the system by using the transfer learning domain. The goal is to provide an end-to-end and high-level deep learning prediction system for clinicians. Or seek medical attention for a computer-aided tool for prognosis prediction and treatment strategies for metastatic brain tumor diseases. The focus of technology is on convenience and generalization.
The first technical highlight is the personalized tumor annotation system, which includes a strong label supervised deep learning model to provide initial tumor markers and assist in personalized tumor screening and bordering through unsupervised deep learning classification. Modifications, the goal of this personal tumor labeling system is to meet the needs of each professional physician.
The second technical highlight is the use of deep network self-encoders and identification, integration of magnetic resonance imaging image factor and structured clinical data, providing integrated prediction models for different dimensional data, and verifying this prediction with more than two hundred cases. Efficacy.
The third technical highlight is the provision of Domain Learning (Transmission Learning) module to overcome the standardization of different MRI images and different medical fields to provide generalization of annotation systems and prognostic systems. Sexual problem solution.
These three technical highlights are integrated into a user-friendly user interface in a one-click form, and will provide end users with a low-computer device experience in the future based on the Edge Computing approach and spirit. It is hoped that the clinical side of the precise treatment of brain metastatic cancer patients can be directly and effectively provided, and it is expected to expand to other cancer treatments.