Method and system for detecting pneumothorax

Inventors

Kim, Min Chul • PARK, Chang Min • Hwang, Eui Jin

Assignees

Lunit Inc • Seoul National University Hospital

Abstract

Some embodiments of the present disclosure provide a pneumothorax detection method performed by a computing device. The method may comprise obtaining predicted pneumothorax information, predicted tube information, and a predicted spinal baseline with respect to an input image from a trained pneumothorax prediction model; determining at least one pneumothorax representative position for the predicted pneumothorax information and at least one tube representative position for the predicted tube information, in a prediction image in which the predicted pneumothorax information and the predicted tube information are displayed; dividing the prediction image into a first region and a second region by the predicted spinal baseline; and determining a region in which the at least one pneumothorax representative position and the at least one tube representative position exist among the first region and the second region.

Core Innovation

The invention relates to a pneumothorax detection method performed by a computing device that receives a medical image and obtains predicted pneumothorax information and predicted tube information from the medical image based on a trained pneumothorax prediction model. The method classifies the medical image based on a relationship between a region in which the pneumothorax exists and tube inclusion in that region.

The method classifies the medical image as an emergency pneumothorax in response to determining that a region in which the pneumothorax exists does not include a tube based on the predicted pneumothorax information and the predicted tube information. The method also supports classification as a general pneumothorax when the pneumothorax region includes the tube.

To support region determination for emergency classification, the approach uses a predicted spinal baseline to divide the image into left and right regions, and determines pneumothorax representative positions and tube representative positions for those regions. The representative position determination uses selection logic including maximum or central value, or threshold-contour-based point selection defined with a threshold and contour connecting points whose prediction value is greater than or equal to a threshold value.

The document further describes training-data generation and model training for the pneumothorax prediction model, including spine-region prediction, spinal baseline extraction, and generation of left/right determination labels for tasks such as pneumothorax prediction, tube prediction, and spinal baseline prediction. The resulting system can provide predicted pneumothorax information and predicted tube information in a visual form such as a prediction heat map and a contour-based representation, and it can trigger an alarm related to tube insertion treatment for the specified region.

Claims Coverage

The independent claims cover a computing-device pneumothorax detection method and a computing device implementing that method, defining a classification rule based on whether a predicted pneumothorax region includes a predicted tube. Across the independent claims, the core inventive features include trained pneumothorax prediction producing both pneumothorax and tube predictions, and emergency pneumothorax classification based on tube absence in the pneumothorax region.

The independent claims define a computing-device system that uses a trained pneumothorax prediction model to obtain predicted pneumothorax information and predicted tube information and then classifies the medical image as an emergency pneumothorax when the pneumothorax region does not include the tube. Dependent claim coverage further includes general pneumothorax classification and supporting derivations such as spinal baseline and contour/threshold-based representations.

Stated Advantages

Documented Applications