Method for object detection using hierarchical deep learning
A hierarchical deep-learning object detection framework provides a method for identifying objects of interest in high-resolution, high pixel count images, wherein the objects of interest comprise a relatively a small pixel count when compared to the overall image. The method uses first deep-learning model to analyze the high pixel count images, in whole or as a patchwork, at a lower resolution to identify objects, and a second deep-learning model to analyze the objects at a higher resolution to classify the objects.
1. A method comprising:
obtaining a high-resolution whole slide image of a post-birth placenta;
analyzing the whole slide image, using a trained machine learning model, to identify one or more blood vessels in the placenta, the analysis occurring at a lower resolution than the native resolution of the whole slide image;
classifying the identified blood vessels at a higher resolution using a trained machine learning classifier that outputs a latent vector for each classified blood vessel:
aggregating the latent vectors for a predetermined number of classified blood vessels;
pooling the aggregated latent vectors by calculating a maximum or minimum of the data for each node of a feature map of the machine learning classifier;
reducing the dimension on the pooled aggregated latent vectors to produce a reduced dimension latent vector; and
performing a binary classification for each whole slide image based on the latent vector.
2. The method of claim 1 wherein the machine learning model places bounding boxes around the identified blood vessels.
3. The method of claim 2 wherein the machine learning model outputs, with each identified blood vessel, a probability that the identified object is a blood vessel.
4. The method of claim 1 , further comprising:
separating the image into a plurality of smaller image patches;
wherein analyzing the comprises analyzing each of the plurality of smaller image patches separately.
5. The method of claim 1 further comprising, prior to the binary classification:
concatenating the reduced dimension latent vector with a vector representing metadata related to the whole slide image;
wherein the reduced dimension latent vector is produced using principal component analysis on the pooled aggregated latent vectors.
6. The method of claim 1 wherein the binary classification is performed using a random decision forest.
7. The method of claim 1 wherein the machine learning classifier classifies each blood vessel as being diseased or healthy.
8. The method of claim 5 wherein the metadata includes one or both of patient metadata regarding a patient from whom the placenta was extracted as well as and metadata regarding the placenta.
9. The method of claim 8 , wherein the metadata comprises one or more of a hypertensive disorder of pregnancy, placental weight, diabetes status of the patient, lupus status of the patient, infant birth weight and full-term/preterm delivery status.
10. The method of claim 8 wherein the binary classification is used to predict a risk of problems occurring in future pregnancies of the patient from whom the placenta was extracted.
11. The method of claim 8 wherein the binary classification is used to predict a risk of preeclampsia in future pregnancies of the patient from whom the placenta was extracted.
12. A method comprising:
obtaining one or more whole slide images of a post-birth placenta;
separating the whole slide image into a plurality of image patches;
analyzing each image patch at a lower resolution than the native resolution of the whole slide image using a trained machine learning model, to identify one or more blood vessels in the image patch;
classifying each identified blood vessel in the image patch at a higher resolution, using a trained machine learning classifier;
aggregating latent vectors produced by the machine learning classifier for a predetermined number of classified blood vessels;
pooling the aggregated latent vectors by calculating a maximum or minimum for each node of a feature map of the machine learning classifier; and
performing principal component analysis on the pooled aggregated latent vectors to produce a reduced dimension latent vector.
13. The method of claim 12 further comprising:
concatenating the reduced dimension latent vector with a vector representing metadata related to the whole slide image; and
performing a binary classification for the whole slide image.
14. The method of claim 13 wherein the metadata related to the whole slide image is metadata related to the birth which produced the placenta.
15. The method of claim 14 wherein the binary classification for the whole slide image includes a probability of disease for the portion of the placenta represented in the whole slide image.
16. The method of claim 12 wherein the binary classification is performed using a random decision forest.
17. The method of claim 12 wherein the trained machine learning model uses a ResNet feature extractor.
18. The method of claim 12 wherein trained machine learning model places a bounding box around each identified blood vessel.
19. The method of claim 12 wherein the binary classification is used to predict a risk of problems occurring in future pregnancies of a patient from whom the placenta was extracted.
20. The method of claim 12 wherein the binary classification is used to predict a risk of preeclampsia in future pregnancies of a patient from whom the placenta was extracted.