Methods for processing PET / CT image data and generating image descriptions using a multimodal deep learning model.

Abstract

The invention relates to a method using a Vision-Language Foundation Model (VLM) to automatically analyze and process whole-body PET / CT image data and generate image descriptions in Vietnamese. The invention also proposes a method for creating a dataset that pairs PET / CT images with corresponding image descriptions, for use in training and evaluating deep learning models. Furthermore, the invention proposes a method for architecture and training deep learning models to generate image descriptions from 3D PET / CT input images. These methods are designed to support the creation and standardization of image descriptions, ensuring consistency, reducing processing time, and minimizing errors.