Method and system for purely geometric machine learning based fractional flow reserve

Patent metadata

Field	Value
Publication number	US-10463336-B2
Application number	US-201515508220-A
Country	US
Kind code	B2
Filing date	Nov 16, 2015
Priority date	Nov 14, 2014
Publication date	Nov 5, 2019
Grant date	Nov 5, 2019

Abstract

A method and system for determining hemodynamic indices, such as fractional flow reserve (FFR), for a location of interest in a coronary artery of a patient is disclosed. Medical image data of a patient is received. Patient-specific coronary arterial tree geometry of the patient is extracted from the medical image data. Geometric features are extracted from the patient-specific coronary arterial tree geometry of the patient. A hemodynamic index, such as FFR, is computed for a location of interest in the patient-specific coronary arterial tree based on the extracted geometric features using a trained machine-learning based surrogate model. The machine-learning based surrogate model is trained based on geometric features extracted from synthetically generated coronary arterial tree geometries.

First claim

The invention claimed is:

1. A method for determining a hemodynamic index for one or more locations of interest in coronary arteries of a patient, comprising:
receiving medical image data of the patient;
extracting patient-specific coronary arterial tree geometry of the patient from the medical image data;
extracting geometric features from the patient-specific coronary arterial tree geometry of the patient; and
computing a hemodynamic index for one or more locations of interest in the patient-specific coronary arterial tree using a trained machine-learning based surrogate model and based purely on the extracted geometric features without considering features from patient-specific physiological measurements, the trained machine-learning based surrogate model trained based on geometric features extracted from synthetically generated coronary arterial tree geometries.

2. The method of claim 1 , wherein the trained machine-learning based surrogate model is trained on geometric features extracted exclusively from synthetically generated non-patient-specific coronary arterial tree geometries.

3. The method of claim 1 , wherein extracting geometric features from the patient-specific coronary arterial tree geometry of the patient comprises:
calculating ischemic weights for coronary artery segments based on the patient-specific coronary arterial tree geometry; and
calculating ischemic contribution scores for coronary artery segments based on the ischemic weights.

4. The method of claim 3 , wherein calculating ischemic weights for coronary artery segments based on the patient-specific coronary arterial tree geometry comprises:
individually calculating initial local ischemic weights for each of a plurality of coronary artery segments based on a respective reference radius value calculated for each of the plurality of coronary artery segments;
calculating a global ischemic weight for the patient-specific coronary arterial tree based on the initial local ischemic weights for the plurality of coronary artery segments; and
calculating final local ischemic weights for each of the plurality of coronary artery segments by distributing the global ischemic weight over the plurality coronary artery segments.

5. The method of claim 4 , wherein calculating a global ischemic weight for the patient-specific coronary arterial tree based on the initial local ischemic weights for the plurality of coronary artery segments comprises:
calculating a plurality of global ischemic weight estimates for the patient-specific coronary arterial tree, wherein each of the plurality of global ischemic weight estimates is calculated from initial local ischemic weights of coronary artery segments from a respective one of a plurality of generations of coronary artery segments; and
calculating the global ischemic weight of the patient-specific coronary arterial tree based on the plurality of global ischemic weight estimates.

6. The method of claim 5 , wherein calculating a plurality of global ischemic weight estimates for the patient-specific coronary arterial tree, wherein each of the plurality of global ischemic weight estimates is calculated from initial local ischemic weights of coronary artery segments from a respective one of a plurality of generations of coronary artery segments comprises, for each of the plurality of generations of coronary artery segments:
assigning a weight to each of the coronary artery segments in that generation of coronary artery segments and leaf coronary artery segments with a generation number smaller than that generation of coronary artery segments; and
calculating the estimate for the global ischemic weight of the patient-specific coronary arterial tree as a function of the initial ischemic weights of coronary artery segments in that generation of coronary artery segments and the leaf coronary artery segments with a generation number smaller than that generation of coronary artery segments and the weights assigned to each of the coronary artery segments in that generation of branches and the leaf coronary artery segments with a generation number smaller than that generation of coronary artery segments.

7. The method of claim 5 , wherein calculating the global ischemic weight of the patient-specific coronary arterial tree based on the plurality of global ischemic weight estimates comprises:
calculating the global ischemic weight of the patient-specific coronary arterial tree as a function of the plurality of global ischemic weight estimates and weights corresponding to the plurality of generations of coronary artery segments.

8. The method of claim 4 , wherein calculating final local ischemic weights for each of the plurality of coronary artery segments by distributing the global ischemic weight over the plurality coronary artery segments comprises:
calculating the final local ischemic weights of a plurality of leaf coronary artery segments by distributing the global ischemic weight over the leaf coronary artery segments based on the initial local ischemic weights of the leaf coronary artery segments; and
calculating the final local ischemic weight for each remaining one of the plurality of coronary artery segments as a sum of the final local ischemic weights of leaf segments downstream from that coronary artery segment.

9. The method of claim 3 , wherein calculating ischemic contribution scores for coronary artery segments based on the ischemic weights comprises:
dividing coronary artery segments into non-anomalous portions and anomalous portions;
calculating the ischemic contribution scores for each non-anomalous portion based on a spatially varying radius of the non-anomalous portion and the ischemic weight of the coronary artery segment in which the non-anomalous portion is located; and
calculating ischemic contribution scores for each anomalous portion based on a first product of the ischemic weight of the coronary artery segment in which the anomalous portion is location and a first mathematical operator applied to a spatially varying radius of the anomalous portion and a second product of a squared ischemic weight of the coronary artery segment in which the anomalous portion is located and a second mathematical operator applied to the spatially varying radius of the anomalous portion.

10. The method of claim 9 , wherein extracting geometric features from the patient-specific coronary arterial tree geometry of the patient further comprises:
for each of the one or more locations of interest, calculating one or more of a cumulative ischemic contribution score from all coronary segments lying between a root segment and a current location, a cumulative ischemic contribution score from the non-anomalous portions of coronary artery segments lying between the root segment and the current location, a cumulative ischemic contribution score from the anomalous portions of coronary artery segments lying between the root segment and the current location, a cumulative ischemic contribution score from all coronary artery segments lying between the current location and a leaf segment, a cumulative ischemic contribution score from the non-anomalous portions of coronary artery segments lying between the current location and a leaf segment, or a cumulative ischemic contribution score from the anomalous portions of coronary artery segments lying between the current location and a leaf segment.

11. The method of claim 3 , wherein extracting geometric features from the patient-specific coronary arterial tree geometry of the patient further comprises:
extracting a plurality of geometric measurements for one or more stenosis regions in the patient-specific coronary arterial tree geometry of the patient.

Assignees

• Siemens Healthcare Gmbh

Inventors

• Itu Lucian Mihai
• Sharma Puneet
• Rapaka Saikiran
• Passerini Tiziano
• Schöbinger Max
• Schwemmer Chris
• Comaniciu Dorin
• Redel Thomas

Classifications

• G06V10/776

  Validation; Performance evaluation · CPC title

• G06F18/2413

  based on distances to training or reference patterns · CPC title

• G06F18/22

  Matching criteria, e.g. proximity measures · CPC title

• G06F18/217

  Validation; Performance evaluation; Active pattern learning techniques · CPC title

• G06V10/42

  Global feature extraction by analysis of the whole pattern, e.g. using frequency domain transformations or autocorrelation · CPC title