Systems and methods for AI-assisted analysis of primary tumor images for prediction of metastases
AssigneesProgenics Pharmaceuticals, Inc., EXINI Diagnostics AB
InventorsKarl Vilhelm Sjöstrand, Kamil Rzechowski
Life of this patent
Estimating the term…
Abstract
Presented herein are systems and methods for the use of automatically-identified suspect regions (e.g., hotspots) within the prostate as imaged in one or more medical images (e.g., PET, CT, or PET/CT image(s)) to predict metastases (e.g., to predict whether localized disease has developed or will develop into metastatic cancer) using a convolutional neural network (CNN).
Claims29
A method for automatically processing one or more medical images of a subject and using the processed image(s) to automatically predict a presence and/or a risk of metastases, the method comprising:
- (a) receiving, by a processor of a computing device, one or more medical images of a prostate of the subject, wherein the one or more medical image(s) comprise a 3D functional image acquired following administration to the subject of an imaging agent;
- (b) automatically identifying, by the processor, within the 3D functional image, a prostate volume corresponding to the prostate of the subject;
- (c) automatically identifying, by the processor, one or more hotspots within the prostate volume, said one or more hotspots corresponding to localized regions of high intensity relative to their surroundings and representing lesions or potential lesions within the subject;
- and
- (d) predicting, by the processor, using a neural network, (i) a presence of metastases in the subject, and/or (ii) a risk of metastases in the subject, said predicting based at least in part on the automatically identified prostate volume and the automatically identified one or more hotspot(s) within the prostate volume, wherein the neural network receives at least two channels of input, the at least two channels of input comprising:
- (A) a prostate intensity channel comprising intensities of voxels located within the prostate volume;
- and
- (B) a hotspot mask channel comprising a mask identifying the one or more hotspots.
A system for automatically processing one or more medical images of a subject and using the processed image(s) to automatically predict a presence and/or a risk of metastases, the system comprising:
- a processor of a computing device;
- and
- memory having instructions stored thereon, wherein the instructions, when executed by the processor, cause the processor to:
- (a) automatically identify, within a 3D functional image acquired following administration to the subject or an imaging agent, a prostate volume corresponding to a prostate of the subject;
- (b) automatically identify one or more hotspots within the prostate volume, said one or more hotspots corresponding to localized regions of high intensity relative to their surroundings and representing lesions or potential lesions within the subject;
- and
- (c) predict, using a neural network, (i) a presence of metastases in the subject, and/or (ii) a risk of metastases in the subject, said predicting based at least in part on the automatically identified prostate volume and the automatically identified one or more hotspots within the prostate volume, wherein the neural network receives at least two channels of input, the at least two channels of input comprising:
- (A) a prostate intensity channel comprising intensities of voxels located within the prostate volume;
- and
- (B) a hotspot mask channel comprising a mask identifying the one or more hotspots.
A method for automatically processing medical images of a subject presenting with localized prostate cancer and using the processed image(s) to automatically predict a risk that the subject will develop metastases, the method comprising:
- (a) receiving, by a processor of a computing device, a medical image of the subject, comprising a graphical representation of a prostate of the subject;
- (b) automatically identifying, by the processor, within the medical image, a prostate volume corresponding to a prostate of the subject;
- (c) automatically identifying, by the processor, one or more hotspots within the prostate volume, said one or more hotspots corresponding to localized regions of high intensity relative to their surroundings and representing lesions or potential lesions within the subject;
- and
- (d) predicting, by the processor, using a neural network, a risk of metastases in the subject based at least in part on the prostate volume and the automatically identified one or more hotspots within the prostate volume,
- wherein the neural network is a trained neural network, having been trained using a plurality of example images each obtained from a particular individual and comprising a graphical representation of hotspots within a prostate region of the particular subject, said plurality of example images comprising:
- (A) a plurality of positive example images obtained for individuals having synchronous metastases;
- and
- (B) a plurality of negative example images obtained for individuals having localized disease.
Dependencies are inferred from the claim text.