Systems and methods for review of computer-aided detection of pathology in images
Disclosed and described herein are systems and methods of performing computer-aided detection (CAD)/diagnosis (CADx) in medical images and comparing the results of the comparison. Such detection can be used for treatment plans and verification of claims produced by healthcare providers, for the purpose of identifying discrepancies between the two. In particular, embodiments disclosed herein are applied to identifying dental caries (“caries”) in radiographs and comparing them against progress notes, treatment plans, and insurance claims.
1. A method of performing computer-aided detection (CAD) of pathologies using an image, comprising:
segmenting each of the one or more teeth that comprise an image to determine boundaries between each of the one or more teeth, wherein each tooth comprises a segmented image;
numbering each of the one or more teeth that comprise the image; and
determining whether each of the one or more teeth that comprise the image has a one or more pathologies using a pathology classifier, wherein the pathology classifier comprises a CNN specifically trained to determine whether a tooth has the one or more pathologies,
wherein segmenting each of the one or more teeth that comprise the image comprises separating upper teeth from lower teeth in the image by identifying an occlusal plane, wherein the occlusal plane is identified by incrementally rotating the image in a clockwise and a counterclockwise direction, where at each of the increments a projection of intensity for each image row is summed and the occlusal plane is identified by a valley in the profiles of this projection, said projections of intensity stored in a column corresponding to the increment;
identifying a maximum valley depth of the occlusal valley in each column, and create an occlusal curve, wherein the occlusal curve is a smoothed curve of the maximum valley depth locations spanned along rows of the image;
identifying boundaries between each tooth or partial tooth that comprises the image, wherein the boundaries are determined independently for an upper and lower set of teeth as determined by the occlusal plane, and wherein the boundaries between each tooth are determined by taking a column-wise projection of intensities between the occlusal curve and an edge of the image, wherein spaces between the teeth are indicated by the valleys in the intensity projections; and
using K-means to determine an average position of column numbers in the image that correspond to the interproximal space between teeth.
2. The method of claim 1 , wherein at least one of the one or more pathologies comprise caries.
3. The method of claim 1 , wherein the image comprises an x-ray image.
4. The method of claim 1 , wherein the image is processed using homomorphic filtering to normalize a range of image intensity throughout the image.
5. The method of claim 1 , further comprising performing CAD on each of the one or more teeth that comprise the image using a convolutional neural network (CNN) classifier model to identify teeth with restorations that may produce false negative results from a pathology classifier.
6. The method of claim 5 , wherein the method further comprises classification of overall image features using the classifier model, wherein the classifier model comprises an image classifier that is trained to identify images that include orthodontia or images that are of poor quality.
7. The method of claim 6 , wherein the classifier model comprises a CNN using the AlexNet architecture or a CNN using the GoogLeNet architecture.
8. The method of claim 5 , wherein performing CAD classification on each of the one or more teeth that comprise the image comprises further scaling the image to a size required by the CNN classifier model, wherein relative physical dimensions of a tooth are preserved relative to any other.
9. The method of claim 1 , further comprising:
starting at a row on the occlusal curve that roughly separates a tooth from one next to it, moving a rectangular window towards the edge of the image, wherein a tooth segmentation line is chosen to be at a column of this window for which the average intensity in the window is the minimum;
filtering out false tooth segmentation curves in the column-wise projections by comparing image intensity along the curve to those of curves translated to the left and right of the segmentation curve, wherein segmentation curves that run through the pulp of a tooth will have more similar intensities to each other, compared to a segmentation curve that runs between the teeth;
determining a distance between intensities, wherein curves for which the intensities are too close to each other are rejected; and
outputting a set of small images, one for each segmented region.
10. The method of claim 1 , further comprising:
performing a fuzzy-logic process by which the teeth can be numbered; and
identifying a likelihood that the segmented tooth image contains any of several types of tooth including primary molars, secondary molars, primary canines, secondary canines, secondary premolars, primary incisors, secondary incisors, gaps between teeth, exfoliating teeth, and cropped teeth, wherein the probabilities of tooth types for the teeth are arranged in a list of length number of teeth multiplied by a number of tooth types,
and wherein the tooth probability list is element-wise multiplied against each row of the matrix to produce a sequence score and tooth labeling is determined by the row with the highest score.
11. The method of claim 1 , wherein an output of the pathology classifier comprises a binary output: “has the one or more pathologies” or “does not have the one or more pathologies 38 .
12. The method of claim 11 , further comprising putting an identifier in the image to identify any of the one or more teeth with the one or more pathologies.
13. The method of claim 1 , further comprising comparing the CAD to procedures performed as described in an insurance claim.
14. The method of claim 13 , wherein the comparison is used by a Payer and the insurance claim is paid in part or in full based on the comparison or the insurance claim is denied in part or in full based on the comparison.
15. The method of claim 13 , wherein any discrepancies between the insurance claim and the CAD are identified and such discrepancies are provided to a dental or healthcare professional that submitted the insurance claim.
16. The method of claim 1 , wherein the CAD is compared to procedures prescribed by a dental or healthcare professional.
17. The method of claim 1 , wherein the CAD is used by a dental or healthcare professional to treat a patient.
18. The method of claim 1 , wherein the CAD is used to audit a dental or healthcare professional.
19. A method of performing computer-aided detection (CAD) of pathologies using an image of one or more teeth, comprising:
segmenting each of the one or more teeth that comprise the image to determine boundaries between each of the one or more teeth, wherein each tooth comprises a separate segmented image;
numbering each tooth of the separate segmented images; and
determining whether each of the teeth that comprise each of the separate segmented images has one or more pathologies using a pathology classifier, wherein the pathology classifier comprises a CNN specifically trained to determine whether a tooth has the one or more pathologies,
wherein numbering each tooth of the separate segmented images comprises:
performing a fuzzy-logic process for the numbering of each tooth in the separate segmented images comprising:
identifying a probability that each of the separate segmented images contains any of a plurality of tooth types, wherein the probabilities of tooth types for the teeth are arranged in a tooth probability list of length number of teeth multiplied by a number of the tooth types,
and wherein the tooth probability list is element-wise multiplied against each row of a matrix to produce a sequence score and tooth numbering is determined by the row with a highest score.