3d segmentation of lesions in ct images using self-supervised pretraining with augmentation
AssigneeMerck Sharp & Dohme Corp.
InventorsYiqiao Liu, Antong Chen, Gregory Goldmacher
Life of this patent
Estimating the term…
Abstract
A method or system for training a convolutional neural network (CNN) for medical imaging analysis. The system pre-trains the CNN's encoder using a dataset of unlabeled 3D medical images. Each 3D image includes an annotated slice delineating a boundary of a lesion and multiple non-annotated 2D slices above and below the annotated slice. The system then fine-tunes the pre-trained encoder using an annotated 2D image dataset. The annotated 2D image dataset includes multiple 2D slices of lesions, each including an annotation that delineates a boundary of a corresponding lesion.
Claims20
A computer-implemented method of training a convolutional neural network (CNN), the method comprising:
- pre-training via contrastive learning, an encoder of the CNN using an unlabeled 3D image dataset, the unlabeled 3D image dataset containing a plurality of 3D medical images, each of which includes an annotated slice and multiple non-annotated 2D slices above and below the annotated slice, the annotated 2D slice including an annotation that delineates a boundary of a lesion;
- and
- fine-tuning the pre-trained encoder using an annotated 2D image dataset, the annotated 2D image dataset containing a plurality of 2D slices of lesions, each of which includes an annotation that delineates a boundary of a corresponding lesion.
A non-transitory computer readable storage medium comprising instructions that, for training of a convolutional neural network (CNN) in multiple stages, the instructions, when executed by one or more processors, causing the one or more processors to:
- pre-train via contrastive learning, an encoder of the CNN using an unlabeled 3D image dataset, the unlabeled 3D image dataset containing a plurality of 3D medical images, each of which includes an annotated slice and multiple non-annotated 2D slices above and below the annotated slice, the annotated 2D slice including an annotation that delineates a boundary of a lesion;
- and
- fine-tune the pre-trained encoder using an annotated 2D image dataset, the annotated 2D image dataset containing a plurality of 2D slices of lesions, each of which includes an annotation that delineates a boundary of a corresponding lesion.
A computer system, comprising:
- one or more processors;
- and
- a memory comprising instructions for training of a convolutional neural network (CNN) in multiple stages, the instructions, when executed by the one or more processors, causing the one or more processors to:
- pre-train via contrastive learning, an encoder of the CNN using an unlabeled 3D image dataset, the unlabeled 3D image dataset containing a plurality of 3D medical images, each of which includes an annotated slice and multiple non-annotated 2D slices above and below the annotated slice, the annotated 2D slice including an annotation that delineates a boundary of a lesion;
- and
- fine-tune the pre-trained encoder using an annotated 2D image dataset, the annotated 2D image dataset containing a plurality of 2D slices of lesions, each of which includes an annotation that delineates a boundary of a corresponding lesion.
Dependencies are inferred from the claim text.