Loading...

quantusOS

ANALYSIS AND CLASSIFICATION OF REDIOGRAPHS FOR THE AUTOMATIC DETECTION OF OSTEOPOROSIS

From a hip X-ray in AP (anteroposterior) view, being valid the whole hip view (two femurs) or half hip (one femur), it is able to detect the risk of osteoporosis. Its technology is based on the quantitative analysis of femur texture.

Classification system applied in quantusOS

  • Class 1: >98.15% Specificity
  • Class 2: >90.74% Specificity
  • Class 3: >55.56% Specificity
  • Class 4: >90.95% Sensitivity
  • Class 5: >98.05% Sensitivity

The system determines, from several thresholds each with values of sensitivity, specificity and positive and negative predictive values shown below, the probability of a sample being positive as a function of the false positives detected in the training sample.

For example, an image classified as "Class 5" is considered a safe positive since, with the chosen threshold, the occurrence of false positives (negative specimens that are classified as positive) is 2% (specificity 98%), so that the positives predicted by the system will be true positives. In this case the number of false negatives (positive samples that are classified as negative) is higher than in the other classes of lower malignancy (low sensitivity).

If an image does not classified as "Class 5", the system checks if it belongs to a class of lesser malignancy, which would be "Class 4".

A specimen classified as "Class 4" is considered to have a high probability of malignancy since, with the chosen threshold, the occurrence of false positives (negative specimens that are classified as positive) is less than 10% (specificity > 90%), so that most of the positives predicted by the system will be true positives. In this case we do have that the number of false negatives (positive samples that are classified as negative) is higher than in other classes of lower malignancy, but lower than in "Class 5" which is of higher malignancy (sensitivity not so low).

If the image is also not classified as "Class 4", the process is repeated for each of the classes of lesser malignancy, until "Class 1" is reached, which is considered benign.

"Class 1" is considered benign for the same argument as explained for "Class 5", but using the values of sensitivity (at 98%) and negative predictive value (lower specificity).

Therefore, the system determines the highest malignancy class to which the image to be analyzed belongs and is indicated in the corresponding report.

Comparative table of quantusOS reliability for each category

Class 1 Class 2 Class 3 Class 4 Class 5
Global Sens / Spec * 98.05% / 24.07% 90.95% / 55.56% 55.99% / 90.74% 31.34% / 98.15%
Sens (Osteopenia vs Osteoporosis) 100.0% / 96.55% 97.44% / 85.96% 76.28% / 40.39% 50.96% / 16.26%
VPP / VPN ** 89.57% / 65.00% 93.15% / 48.00% 97.57% / 23.67% 99.11% / 17.69%
* Sens: Sensitivity Spec: Specificity
** VPP: Positive Prediction Value VPN: Negative Prediction Value

WHY DOES quantusOS WORK?

quantusOS is presented as a novel Artificial Intelligence method based on state-of-the-art Deep Learning, with 3 different classification algorithms: such as dividing images into single femurs, excluding femurs with prostheses and cropping the images by the upper part of the femur, in order to improve the training of the prediction algorithm.

Its technology is based on performing a quantitative analysis of the radiograph. This analysis makes it possible to identify patterns, thus assisting in the automatic identification of osteoporosis.

quantusOS:



Non-invasive


Reliable


Fast