Method for label-free imaging and classification of malaria parasites
A method of measuring malarial parasitemia includes disposing a sample including red blood cells in liquid form on a sample stage, illuminating the sample with optical radiation, capturing a plurality of images of the sample, and extracting, from the one or more of the plurality of images, a set of red blood cell images. Each red blood cell image is associated with a particular red blood cell. The method also includes for each red blood cell image in the set of red blood cell images, inputting each red blood cell image into a machine learning model and generating, using the machine learning model, a classification related to a malaria parasite lifecycle stage for each of the red blood cells. The method further includes determining the malarial parasitemia for the sample.
1 . A method of measuring malarial parasitemia, the method comprising:
disposing a sample including a plurality of red blood cells in liquid form on a sample stage;
illuminating the sample with optical radiation;
capturing, at a first focal plane, first image of the plurality of red blood cells in the sample;
extracting, from the first image, a first red blood cell image associated with a particular red blood cell;
capturing, at a second focal plane, a second image of the plurality of red blood cells in the sample;
extracting, from the second image, a second red blood cell image associated with the particular red blood cell;
computing a first focus metric for the first image;
computing a second focus metric for the second image;
forming an input set by selecting the first image if the first focus metric is greater than the second focus metric or the second image if the second focus metric is greater than the first focus metric;
inputting the input set into a machine learning model; and
generating, using the machine learning model, a classification related to a malaria parasite lifecycle stage for the particular red blood cell; and
determining the malarial parasitemia for the sample.
2 . The method of claim 1 wherein the classification related to the malaria parasite lifecycle stage comprises a suite of confidence scores, wherein each of the confidence scores in the suite of confidence scores defines a probability that the particular red blood cell is associated with one of a plurality of malaria parasite lifecycle stages.
3 . The method of claim 2 wherein the suite of confidence scores comprises:
a first confidence score associated with a ring-stage parasite;
a second confidence score associated with a trophozoite-stage parasite; and
a third confidence score associated with a schizont-stage parasite.
4 . The method of claim 2 wherein the suite of confidence scores comprises:
a first confidence score associated with a malaria-free red blood cell;
a second confidence score associated with a ring-stage parasite;
a third confidence score associated with a trophozoite-stage parasite; and
a fourth confidence score associated with a schizont-stage parasite.
5 . The method of claim 4 further comprising summing:
a number of cells for which the third confidence score is the highest confidence score in the suite of confidence scores; and
a number of cells for which the fourth confidence score is the highest confidence score in the suite of confidence scores; wherein
the malarial parasitemia is a late-stage parasitemia.
6 . The method of claim 4 wherein the malarial parasitemia is an overall parasitemia computed by:
determining a number of cells that are parasitemic as a sum of cells for which the second confidence score, the third confidence score, or the fourth confidence score is the highest confidence score in the suite of confidence scores for the cell; and
dividing the number of cells that are parasitemic by a total number of cells.
7 . The method of claim 1 wherein the sample stage comprises a flow cell.
8 . The method of claim 1 wherein the sample comprises live cells.
9 . The method of claim 1 further comprising, prior to extracting the first red blood cell image, performing semantic segmentation on the first image.
10 . The method of claim 1 wherein extracting the first red blood cell image and the second red blood cell image comprises determining a bounding box associated with each red blood cell image.
11 . The method of claim 1 wherein the malarial parasitemia is categorized as an early stage parasitemia associated with a ring-stage parasite or a late-stage parasitemia associated with a trophozoite-stage or a schizont-stage parasite.
12 . The method of claim 1 wherein the optical radiation comprises light in the ultraviolet light wavelength range.
13 . The method of claim 1 wherein the optical radiation comprises light having a wavelength between 350 nm and 420 nm.
14 . The method of claim 13 wherein the wavelength is 365 nm.
15 . The method of claim 13 wherein the wavelength is 405 nm.
16 . The method of claim 1 wherein the sample comprises purified red blood cells immersed in liquid.
17 . The method of claim 1 wherein the sample comprises whole blood.
18 . The method of claim 1 wherein the sample is free of Giemsa staining.
19 . The method of claim 1 wherein the first red blood cell image and the second red blood cell image are each associated with a different wavelength.