US 11,810,299 B1
Method and system for computer aided diagnosis using ensembled 2D and 3D neural networks based on medical images
Benoît Huet, Roquefort-les Pins (FR); Danny Francis, Antibes (FR); and Pierre Baudot, Marseilles (FR)
Assigned to MEDIAN TECHNOLOGIES, Valbonne (FR)
Filed by MEDIAN TECHNOLOGIES, Valbonne (FR)
Filed on Jul. 13, 2023, as Appl. No. 18/351,699.
Claims priority of application No. 22315156 (EP), filed on Jul. 15, 2022.
Int. Cl. G06T 7/00 (2017.01); G06V 10/25 (2022.01); G06V 10/26 (2022.01); G06V 10/774 (2022.01); G06V 10/82 (2022.01); G06T 7/73 (2017.01); G06V 10/32 (2022.01); G06V 10/776 (2022.01); G16H 50/20 (2018.01)
CPC G06T 7/0012 (2013.01) [G06T 7/73 (2017.01); G06V 10/25 (2022.01); G06V 10/26 (2022.01); G06V 10/32 (2022.01); G06V 10/774 (2022.01); G06V 10/776 (2022.01); G06V 10/82 (2022.01); G16H 50/20 (2018.01); G06T 2200/04 (2013.01); G06T 2207/10081 (2013.01); G06T 2207/10088 (2013.01); G06T 2207/20081 (2013.01); G06T 2207/20084 (2013.01); G06T 2207/30096 (2013.01); G06V 2201/03 (2022.01)] 27 Claims
OG exemplary drawing
 
1. A method for generating a neural network for characterizing a plurality of Regions Of Interest ROIs based on a plurality of 3D medical images, comprising the steps of:
Providing said plurality of 3D medical images and
Localizing a plurality of ROIs in said plurality of 3D medical images, and for each ROI, defining a 3D bounding box by a set of 6 extreme coordinates (Xmin,Xmax,Ymin,Ymax,Zmin,Zmax), and
On one hand:
For each defined 3D bounding box, 2D preprocessing in order to extract a plurality of Ki 2D patches from each ROI comprised in said 3D bounding box, and
Training a 2D neural network based on the extracted 2D patches,
On the other hand:
For each defined 3D bounding box, 3D preprocessing in order to extract at least one 3D patch from each ROI comprised in said 3D bounding box, and
Training a 3D neural network for characterizing said plurality of ROIs based on the extracted 3D patches for the plurality of ROIs,
wherein the step of 2D preprocessing comprises at least extracting a plurality of Ki 2D patches as contiguous and parallel slices with respect to a single predetermined plane, and
wherein the 2D neural network is configured to be trained for characterizing the plurality of ROIs based on the extracted plurality of 2D patches for the plurality of ROIs.