Method and system for monitoring and evaluation of pressure ulcer severity

1. A method for evaluating a pressure ulcer, comprising:

a pre-processing step of detecting in a color input image a boundary of the pressure ulcer to get a RGB region of interest (ROI) image;

converting the RGB ROI image to a grayscale ROI image;

processing the grayscale ROI image using a Linear Combination of Discrete Gaussians (LCDG) process to estimate three main class probabilities of the grayscale ROI image;

processing the RGB ROI image using predetermined RGB values to estimate the three main class probabilities of the RGB ROI image;

combining the estimated three main class probabilities of the grayscale ROI image and the estimated three main class probabilities of the RGB ROI image to determine an estimated labeled image; and

normalizing and refining the estimated labeled image using a Generalized Gauss-Markov Random Field (GGMRF) process to produce a final segmentation image.

2. The method of claim 1 , wherein processing the grayscale ROI image using the LCDG process includes:

collecting a marginal empirical probability distribution Fs=(fs(q): q∈Q) of gray levels;

finding an initial LCDG-model that closely approximates Fs by using an initializing process to estimate numbers C p −K, C n , and parameters w, Θ (weights, means, and variances) of positive and negative Discrete Gaussians (DGs);

refining the initial LCDG-model with the numbers C p and C n by adjusting all other parameters with an expectation-maximization (EM) method to create a final LCDG-model;

splitting the final LCDG-model into K sub-models, one per each dominant mode, by minimizing expected errors of misclassification; and

selecting from the K sub-models a sub-model with a largest mean value as a desired model.

3. The method of claim 2 , wherein the steps of finding the initial LCDG-model that closely approximates Fs and refining the initial LCDG-model produce a LCDG-model with nonnegative starting probabilities p w,Θ (q), and the steps of splitting the final LCDG-model into K sub-models and selecting from the K sub-models a sub-model with the largest mean value increase a likelihood that the probabilities continue to be nonnegative, such upon completion of these steps, three probabilities for each pixel in the grayscale ROI image are obtained: P 1 which is the probability to be belonging to class 1, P 2 which is the probability to be belonging to class 2, and P 3 which is the probability to be belonging to class 3.

4. The method of claim 1 , wherein processing the RGB ROI image using predetermined RGB values includes:

constructing a database for a set of labeled images for three main tissue classes;

constructing a matrix for each class by aggregating RGB values of all pixels of that class from training images which have been labeled;

for each pixel in the RGB ROI image, calculating Euclidean distances between the pixel and the matrix for each class;

determining for each of the three main tissue classes a total number of voxels that are at distances lower than a predefined threshold; and

estimating the three main class probabilities for each pixel in the RGB ROI image using individual probabilities.

5. The method of claim 1 , wherein combining the estimated three main class probabilities of the grayscale ROI image and the estimated three main class probabilities of the RGB ROI image includes:

constructing a final pixel-wise probability vector as P=P Gray ×P Color ; and

obtaining the estimated labeled image as an initial tissue classification of each pixel using a vector element with a highest probability.

6. A method of monitoring a pressure ulcer, comprising determining an objective change in the pressure ulcer by comparing current results with prior results of the method of evaluating the pressure ulcer of claim 1 .

7. A system for monitoring and evaluating a pressure ulcer, comprising:

a microprocessor;

a data storage device in communication with the microprocessor;

an input/output device in communication with the microprocessor; and

a digital camera in communication with the microprocessor;

the microprocessor for executing instructions stored in the data storage device for:

receiving a color input image of the pressure ulcer from the digital camera;

detecting in the color input image a boundary of the pressure ulcer to get a RGB region of interest (ROI) image;

converting the RGB ROI image to a grayscale ROI image;

processing the grayscale ROI image using a Linear Combination of Discrete Gaussians (LCDG) process to estimate three main class probabilities of the grayscale ROI image;

processing the RGB ROI image using predetermined RGB values to estimate the three main class probabilities of the RGB ROI image;

combining the estimated three main class probabilities of the grayscale ROI image and the estimated three main class probabilities of the RGB ROI image to determine an estimated labeled image;

normalizing and refining the estimated labeled image using a Generalized Gauss-Markov Random Field (GGMRF) process to produce a final segmentation image; and

outputting the final segmentation image to the input/output device.

8. The system of claim 7 , wherein processing the grayscale ROI image using the LCDG process includes:

collecting a marginal empirical probability distribution Fs=(fs(q): q∈Q) of gray levels;

finding an initial LCDG-model that closely approximates Fs by using an initializing process to estimate numbers C p −K, C n , and parameters w, Θ (weights, means, and variances) of positive and negative Discrete Gaussians (DGs);

refining the initial LCDG-model with the numbers C p and C n by adjusting all other parameters with an expectation-maximization (EM) method to create a final LCDG-model;

splitting the final LCDG-model into K sub-models, one per each dominant mode, by minimizing expected errors of misclassification; and

selecting from the K sub-models a sub-model with a largest mean value as a desired model.

9. The system of claim 8 , wherein the steps of finding the initial LCDG-model that closely approximates Fs and refining the initial LCDG-model produces a LCDG-model with nonnegative starting probabilities p w,Θ (q), and the steps of splitting the final LCDG-model into K sub-models and selecting from the K sub-models a sub-model with the largest mean value increases a likelihood that the probabilities continue to be nonnegative, such that upon completion of these steps, three probabilities for each pixel in the grayscale ROI image are obtained: P 1 which is the probability to be belonging to class 1, P 2 which is the probability to be belonging to class 2, and P 3 which is the probability to be belonging to class 3.

10. The system of claim 7 , wherein processing the RGB ROI image using predetermined RGB values includes:

constructing a database for a set of manually labeled images for three main tissue classes;

constructing a matrix for each class by aggregating RGB values of all pixels of that class from training images which have been manually labeled by an expert;

for each pixel in the RGB ROI image, calculating Euclidean distances between the pixel and the matrix for each class;

determining for each of the three main tissue classes a total number of voxels that are at distances lower than a predefined threshold; and

estimating the three main class probabilities for each pixel in the RGB ROI image using individual probabilities.

11. The system of claim 7 , wherein combining the estimated three main class probabilities of the grayscale ROI image and the estimated three main class probabilities of the RGB ROI image includes:

constructing a final pixel-wise probability vector as P=P Gray ×P Color ; and

obtaining the estimated labeled image as an initial tissue classification of each pixel using a vector element with a highest probability.

12. The system of claim 7 , further comprising a power supply in communication with the microprocessor, the data storage device, the input/output device, and the digital camera; wherein the integrated power supply, the microprocessor, the data storage device, the input/output device, and the digital camera are integrated into a unitary device that can be carried by a caregiver and utilized at a bedside of a patient.