Generating lesion change prediction data for medical scans

Patent metadata

Field	Value
Publication number	US-11694136-B2
Application number	US-202217656337-A
Country	US
Kind code	B2
Filing date	Mar 24, 2022
Priority date	Nov 21, 2018
Publication date	Jul 4, 2023
Grant date	Jul 4, 2023

Abstract

A method includes generating a longitudinal lesion model by performing a training step on a plurality of sets of longitudinal data. Dates of medical scans of different ones of the plurality of sets of longitudinal data have relative time differences corresponding to different time spans, and each set of the plurality of sets of longitudinal data corresponds to one of a plurality of different patients. The longitudinal lesion model is utilized to perform an inference step on a received medical scan to generate, for a lesion detected in the received medical scan, a plurality of lesion change prediction data for a corresponding plurality of different projected time spans ending after the current date. At least one of the plurality of lesion change prediction data is transmitted for display.

First claim

What is claimed is:

1. A method comprising:
generating, in a first temporal period, a longitudinal lesion model by performing a training step utilizing a plurality of sets of longitudinal data, wherein dates of medical scans of different ones of the plurality of sets of longitudinal data have relative time differences corresponding to different time spans, wherein each set of the plurality of sets of longitudinal data corresponds to one of a plurality of different patients, and wherein one set of longitudinal data of the plurality of sets of longitudinal data corresponds a first patient and is determined based on:
a first lesion volume measurement of a first lesion detected in a first medical scan of the one set of longitudinal data; and
a second lesion volume measurement of the first lesion detected in a second medical scan of the one set of longitudinal data, wherein the second medical scan has a more recent scan date than the first medical scan;

receiving, in a second temporal period that is strictly after the first temporal period, a third medical scan that is associated with a second patient that is distinct from the first patient;
utilizing the longitudinal lesion model to perform an inference step on the third medical scan to generate, for a second lesion detected in the third medical scan, a plurality of lesion change prediction data for a corresponding plurality of different projected time spans ending after a current date based on the different time spans; and
transmitting at least one lesion change prediction data of the plurality of lesion change prediction data for display.

2. The method of claim 1 , wherein generating the first lesion volume measurement of the first lesion detected in the first medical scan of the one set of longitudinal data is based on:
detecting the first lesion in the first medical scan by utilizing a computer vision model;
determining a first subset of image slices of a first plurality of image slices of the first medical scan that include the first lesion;
performing a lesion area calculation for each one of the first subset of image slices to generate a first set of lesion area measurements; and
performing a lesion volume calculation on the first set of lesion area measurements to generate the first lesion volume measurement.

3. The method of claim 2 , wherein the first medical scan indicates modality data, wherein performing the lesion volume calculation includes performing a Riemann sum calculation on the first set of lesion area measurements, and wherein the Riemann sum calculation utilizes a uniform partition width determined based on a slice thickness of each of the first plurality of image slices indicated in the modality data.

4. The method of claim 3 , wherein the first subset of image slices are consecutive image slices of the first plurality of image slices, wherein every pair of consecutive image slices of the first subset of image slices corresponds to one of a plurality of trapezoidal areas, wherein performing the lesion volume calculation includes performing a summation of the plurality of trapezoidal areas, wherein each of the plurality of trapezoidal areas is calculated by multiplying the slice thickness by half of the sum of a first base and a second base, wherein a value of the first base is equal to a first one of the first set of lesion area measurements corresponding to a first one of a corresponding pair of consecutive image slices, and wherein a value of the second base is equal to a second of the of the first set of lesion area measurements corresponding to a second one of the corresponding pair of consecutive image slices.

5. The method of claim 2 , wherein generating the first lesion volume measurement of the first lesion detected in the first medical scan of the one set of longitudinal data is further based on:
performing a lesion diameter calculation on at least one of the first subset of image slices to generate a first lesion diameter measurement; and
generating Response Evaluation Criteria in Solid Tumors (RECIST) eligibility data, wherein the RECIST eligibility data indicates an eligibility criterion is met when the first lesion diameter measurement compares favorably to a diameter threshold, and wherein the RECIST eligibility data indicates an eligibility criterion is not met when the first lesion diameter measurement compares unfavorably to the diameter threshold, and
transmitting the RECIST eligibility data for display.

6. The method of claim 5 , further comprising:
receiving human measurement data associated with the first medical scan that includes a prior diameter measurement of the first lesion in the first medical scan performed by a human that previously reviewed the first medical scan;
generating measurement accuracy data by comparing the prior diameter measurement to the first lesion diameter measurement; and
transmitting the measurement accuracy data for display.

7. The method of claim 5 , wherein the lesion diameter calculation is performed on each of the first subset of image slices to generate a set of diameter measurements, and wherein generating the first lesion diameter measurement includes selecting a maximum of the set of diameter measurements.

8. The method of claim 5 , wherein the first lesion diameter measurement corresponds to a segment connecting a first point and a second point of a perimeter of the first lesion in one of the first subset of image slices, and wherein the segment is oblique to an x-axis of the one of the first subset of image slices.

9. The method of claim 8 , wherein generating the first lesion diameter measurement is further based on:
determining a set of pixels of the one of the first subset of image slices that correspond to the perimeter of the first lesion in the one of the first subset of image slices, wherein the set of pixels includes a first pixel corresponding to the first point and a second pixel corresponding to the second point;
calculating a set of segment lengths corresponding to a distance between each of a plurality of pairs of pixels, wherein the plurality of pairs of pixels includes every combination of selecting two of the set of pixels; and
determining the first lesion diameter measurement by selecting a maximum of the set of segment lengths, wherein the maximum of the set of segment lengths corresponds to the distance between the first pixel and the second pixel.

10. The method of claim 5 , further comprising:
performing the lesion diameter calculation on at least one of a second subset of image slices detected to include the first lesion in the second medical scan to generate a second lesion diameter measurement;
determining that the first lesion is a target lesion based on the RECIST eligibility data;
determining RECIST evaluation data by calculating a difference between the first lesion diameter measurement and the second lesion diameter measurement, wherein the RECIST evaluation data indicates one of: Complete Response, Partial Response, Stable Disease, or Progressive Disease; and
transmitting the RECIST evaluation data for display.

11. The method of claim 5 , wherein the first lesion diameter measurement corresponds to a segment connecting a first point on a first one of the first subset of image slices and a second point on a second one of the first subset of image slices.

12. The method of claim 1 , further comprising:
Determining that the first lesion corresponds to a lung nodule;
performing a lesion diameter calculation on at least one of a second subset of image slices detected to include the first lesion in the second medical scan to generate a lesion diameter measurement;
determining a Lung-RADS score for t

Assignees

• Enlitic Inc

Inventors

• Lyman Kevin
• Upton Anthony
• Covington Ben
• Yao Li
• Lui Keith

Classifications

• G06T12/20

  Inverse problem, i.e. transformations from projection space into object space · CPC title

• G06T11/26

  Drawing of charts or graphs · CPC title

• G06T11/10

  Texturing; Colouring; Generation of textures or colours (retouching, inpainting or scratch removal G06T5/77) · CPC title

• G06N3/045

  Combinations of networks · CPC title

• G06N3/09

  Supervised learning · CPC title