Method for Determining a Probability of a Bone Disease in a Patient

Simple SummaryContent extracted with AI.

This invention is a method for estimating the probability that a patient has a bone disease, using a two-stage artificial neural network approach. In a training phase, a first neural network is trained on a large image dataset, and its learned parameters are then transferred to a second neural network — a technique known as transfer learning. The second network is fine-tuned on a more specific bone-disease image dataset and then deployed in a clinical examination phase. During examination, skeletal imaging data from a patient is fed into the second trained model, which outputs a probability score indicating the likelihood of the patient having the bone disease.

Use CasesContent extracted with AI.

• Automated screening for osteoporosis using X-ray or CT scans of the skeleton in radiology departments.
• Early detection of bone metastases in oncology patients by analyzing skeletal imaging data.
• Computer-aided diagnosis support for orthopedic surgeons assessing patients for conditions such as osteoarthritis or Paget's disease.
• Population-level bone disease risk stratification in preventive healthcare programs using routine imaging data.
• Reducing radiologist workload in high-volume imaging centers by pre-filtering scans with a high probability of bone pathology.

BenefitsContent extracted with AI.

• Transfer learning from a large general image dataset to a smaller bone-disease-specific dataset reduces the need for large labeled medical imaging datasets, which are costly and scarce.
• The two-stage training approach improves diagnostic accuracy compared to training a single network solely on limited medical imaging data.
• Automated probability scoring provides consistent, reproducible assessments that are free from inter-observer variability typical of manual radiological review.
• The method enables faster patient triage by delivering near-instant risk probability scores from skeletal images without requiring specialist intervention at the point of screening.
• Leveraging pre-trained network parameters lowers computational training costs and shortens the time required to deploy a clinically usable model.

Technical Classifications (CPCs)

Inventors & Applicants

Patent Abstract

The subject matter of the invention is a method for determining a probability with which a patient has a bone disease (1), comprising the following method steps: in a training phase (11): training a first artificial neural network using a training image dataset, wherein the training image dataset, transferring network model parameters of a first network model (3) formed by training the first artificial neural network to a second artificial neural network, training the second artificial neural network based on the network model parameters using a further training image dataset, providing a second network model (4) formed by training the second artificial neural network, and in an examination phase (12) subsequent to the training phase (11): acquiring image data (5) of the skeletal structure of a patient, and inputting the image data (5) into the second network model (4), wherein the second network model (4) provides, in response to the input of the image data (5), a probability with which the patient has the bone disease (1). In this manner, an improved method for determining a probability with which a patient has a bone disease (1) based on image data (5) of a skeletal structure is provided.