System and method for detecting neovascular lesions or exudative age-related macular degeneration

Abstract

A computer-based system for detecting neovascularization (NV) lesions or exudative age-related macular degeneration (xAMD) is disclosed. The system includes a processor, memory, and storage, and is configured to: receive optical coherence tomography (OCTA) images of four retinal plexuses from a patient: superficial microvascular plexus (SCP), deep microvascular plexus (DCP), outer retinal (OR), and choroidal microvessels (CC); evaluate the OCTA image data using a deep neural network (DNN) model to generate a determination of the NV lesion; subsequently generate a determination of NV exudative activity indicating the xAMD; and provide the results to the patient or a third party. The DNN model includes MobileNet and DenseNet modules. A computer-based method for detecting NV lesions or xAMD is also disclosed.

Claims

1. A computer-implemented system for detecting neovascularization or exudative age-related macular degeneration, the system comprising a processor, memory, and storage, and configured to: receive optical coherence tomography (OCTA) images of four retinal plexuses from a patient: superficial capillary plexus (SCP), deep capillary plexus (DCP), outer retina (OR), and choroidal capillary (CC); evaluate the OCTA images using a deep neural network (DNN) model to generate a determination of the neovascularization, and subsequently generate a determination of neovascular exudative activity indicative of the exudative age-related macular degeneration; and provide the determination to the patient or a third party; wherein the DNN model includes a MobileNet module and a DenseNet module; The system is further configured to perform an input segmentation, suppression, and exchange perturbation (InSSS-P) evaluation to determine the anatomical significance of a selected input for the DNN model.

2. The system according to claim 1, wherein the MobileNet module is configured to extract features of the vascular plexus layer.

3. The system according to claim 1, wherein the DenseNet module is configured to extract vascular features at the branch level.

4. The system according to claim 1, wherein the InSSS-P consists of an input segmentation perturbation, an input suppression perturbation, and an input switching perturbation.

5. A computer-implemented method for detecting neovascularization or exudative age-related macular degeneration, comprising: receiving optical coherence tomography (OCTA) images of four retinal vascular plexuses from a patient: superficial microvascular plexus (SCP), deep microvascular plexus (DCP), outer retinal (OR), and choroidal microvascular plexus (CC); evaluating the OCTA images using a deep neural network (DNN) model to generate a determination of the neovascularization, and subsequently generating a determination of neovascular exudative activity indicative of the exudative age-related macular degeneration; and providing the determination to the patient or a third party; wherein the DNN model includes a MobileNet module and a DenseNet module; wherein the method further comprises: performing an input segmentation, suppression, and exchange perturbation (InSSS-P) evaluation to determine the anatomical significance of the selected input to the DNN model.

6. The method according to claim 6, wherein the MobileNet module is configured to extract features of the vascular plexus layer.

7. The method according to claim 6, wherein the DenseNet module is configured to extract vascular features at the branching level.

8. The method according to claim 5, wherein the InSSS-P consists of an input segmentation perturbation, an input suppression perturbation, and an input switching perturbation.

Images