Predicting lung cancer risk

Inventors

Putha, Preetham • Tadepalli, Manoj • Reddy, Bhargava • Raj, Tarun • Jagirdar, Ammar • Rao, Pooja • Warier, Prashant

Assignees

Qure AI Technologies Pvt Ltd

Abstract

A system and method for predicting a lung cancer risk based on a chest X-ray in which a nodule is detected in a chest of a patient based on an analysis of the chest X-ray using an image processing technique. A region of interest associated with the nodule is identified using the image processing technique. The region of interest is further analyzed using deep learning to determine a plurality of characteristics associated with the nodule. The plurality of characteristics comprises a size of the nodule, a calcification in the nodule, a homogeneity of the nodule and a spiculation of the nodule. Further, the plurality of characteristics is compared with a trained data model using deep learning. Based on the comparison, a risk score associated with the nodule is generated. Further, the lung cancer risk is predicted when the risk score exceeds a predefined threshold value.

Core Innovation

A deep-learning method and system predict a lung cancer risk based on a chest X-ray by receiving a chest X-ray of a patient, detecting a nodule in the chest X-ray using an image processing technique, and identifying a region of interest associated with the nodule as an area surrounding the nodule. The method determines a plurality of characteristics associated with the nodule based on an analysis of the region of interest using deep learning.

The plurality of characteristics comprises a size of the nodule, a calcification in the nodule, a homogeneity of the nodule, and a spiculation of the nodule. The method compares the plurality of characteristics with a trained data model using deep learning, where the trained data model comprises historical data related to different sizes of nodules, concentration of calcification in nodules, information associated with homogeneity of nodules, and spiculations for the different nodules.

The method generates a risk score for the nodule in real time based on the comparison of the plurality of characteristics with the trained data model. The method predicts a lung cancer risk when the risk score for the nodule exceeds a predefined threshold value, and dependent refinements further include monitoring a change in size of the nodule using a previous chest X-ray to map increases and decreases to high and low risk.

Claims Coverage

The document provides three independent claims (method, system, and non-transitory computer program product). Across these independent claims, the core inventive features cover nodule detection and region-of-interest selection on chest X-rays, deep-learning determination of nodule characteristics, comparison to a deep-learning trained model built from historical data, and real-time risk scoring with a predefined threshold decision rule.

Overall claim coverage centers on detecting nodules in chest X-rays, forming a region of interest around the nodule, extracting deep-learning nodule characteristics (size, calcification, homogeneity, spiculation), comparing these characteristics to a deep-learning trained data model based on historical data, and producing a real-time risk score used to predict lung cancer risk when the score exceeds a predefined threshold.

Stated Advantages

Documented Applications