Methods of assessing lung disease in chest x-rays

1. A system for detecting lung abnormalities, the system comprising:

an image pre-processing module that resizes a chest x-ray image to produce a first image at a down-sampled or up-sampled resolution and segments the image into at least one subsection of the image that represents an organ of a body;

a collection of neural networks, trained using a plurality of network architectures, wherein the collection of neural networks includes at least

a first neural network that analyzes the first image, and

a second neural network that analyzes the subsection of the image, wherein the neural networks of the collection each independently make an inference as to the presence of an abnormality;

an ensemble classifier that reports the presence of an abnormality at a location in the lung using the collection of neural network inferences as inputs; and

wherein the system is operable to analyze the first image for quality and positioning accuracy and reject an image from processing and, when rejected, provide a real-time notification instructing a technician to acquire another image.

2. The system of claim 1 , wherein each neural network of the collection is one of Faster R-CNN, Inception-Resnet, DenseNet, or NasNet.

3. The system of claim 1 , wherein the system applies one or more exclusion criteria to the image and optionally wherein the criteria include patient age, over exposure, under exposure, or content of image metadata.

4. The system of claim 1 , wherein the pre-processing module (i) checks for the validity of the image quality and positioning, (ii) standardizes the image brightness and contrast, and/or (iii) standardizes the image across a plurality of images acquisition devices.

5. The system of claim 1 , further comprising a feature engineering module that creates features from the collections of neural networks and provides the features as inputs for the ensemble classifier.

6. The system of claim 1 , wherein the ensemble classifier uses averaging, logistic regression, a generalized linear model, or a random forest.

7. The system of claim 1 , wherein the subsection is at an original first resolution of the chest x-ray image, and wherein the first image is at a lower second resolution than the original resolution, and wherein the collection comprises a third neural network, and wherein the system parses the image into one or more segments that include a third image of intermediate resolution, and wherein the third neural network analyzes the third image to make an inference as to the presence of an abnormality.

8. The system of claim 7 , wherein the second neural network classifies objects by creating a heatmap of bounded potential objects and their corresponding confidence scores and classifies objects using the heatmap.

9. The system of claim 8 , wherein the second neural network further comprises a region proposal network.

10. The system of claim 1 , wherein the system segments the image to select the subsection by performing an object detection operation on the image to create a region proposal for an object detected in the image, and selects the subsection from within the region proposal.

11. The system of claim 10 , wherein second neural network assigns a confidence score to the bounded potential objects.

12. The system of claim 10 , wherein the second neural network classifies potential objects as detected objects using the likelihood score.

13. The system of claim 1 , wherein the second neural network comprises a Fast RCNN.

14. A system for detecting lung abnormalities, the system comprising:

an image pre-processing module that (i) resizes a chest x-ray file to produce a first image at a down-sampled resolution, and (ii) places a subsection of the of the chest x-ray file into a second image at an original resolution of the chest x-ray file;

a first neural network that analyzes the first image to output a first set of scores indicating probabilities of nodules at locations in the lung;

a second neural network that analyzes the second image to output a second set of scores of probabilities of a nodule at a location in the lung;

an ensemble classifier that reports the presence of the nodule in the location in the lung using the first set of scores and the second set of scores as inputs; and

wherein the second neural network performs an object detection operation on the chest x-ray file, creates a bounding box for an object detected in the file, and selects the subsection for the second image from within the box.

15. The system of claim 14 , wherein the detected objects include sub-visual objects.

16. The system of claim 14 , further comprising a third neural network, and wherein the pre-processing module further (iii) parses the chest x-ray file into one or more segments to create at least one third image of intermediate resolution, and wherein the third neural network processes the third image to score for probability of a cancerous nodule in the lung.

17. The system of claim 14 , wherein the neural networks have been trained using chest x-ray images, lung CT scans, lung PET-CT scans, and/or clinical outcome data.

18. A diagnostic method, the method comprising:

providing an image file of a chest x-ray from a patient to a machine learning system that operates by resizing the image file of the chest x-ray into a first image that depicts the entire x-ray at a reduced resolution and placing a subsection of the file into a second image at an original resolution;

analyzing the first and second images in parallel by respective first and second neural networks to output scores indicating a probability of a nodule, wherein the machine learning system has been trained to learn associations between features in chest x-rays and known pathology results with an area under the curve (AUC) of true positives over false positives for learned feature associations is between 0.7 and 1; and

operating the machine learning system to detect lung nodules.