AI Head and Neck Landmark Detection With Two-Stage CT Analysis

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing methods for detecting anatomical landmarks in three-dimensional CT images of the head and neck using artificial intelligence face challenges such as low speed, high cost, and difficulty in learning due to varying image densities from different CT apparatuses, with manual methods being inaccurate and labor-intensive.

Innovation Solution

A two-step process involving conversion of high-capacity CT images to low-resolution images for initial landmark detection, followed by secondary detection in specific areas, using a CNN-based model that learns from images across different apparatuses, maintaining accuracy and speed.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If a CNN landmark detection model is used to detect landmarks from large-capacity three-dimensional CT images, then detection accuracy is improved, but processing speed decreases and computational cost increases

Engineering Contradiction:
Improvelandmark detection accuracyVSAvoiddetection speed
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The patent applies segmentation by dividing the large-capacity three-dimensional CT image into multiple low-resolution CT images. This allows the CNN model to process smaller image segments instead of the entire large image at once, thereby improving processing speed while maintaining detection accuracy through focused analysis of divided regions.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent extracts only the necessary portions of the CT image by converting three-dimensional CT images into multiple low-resolution CT images that contain relevant landmark information. This extraction approach reduces the computational burden on the CNN model while preserving the essential features needed for accurate landmark detection.

Inventive Principle:
Principle #2Taking out (Extraction)

2Measurement precision

If a CNN landmark detection model is used to detect landmarks from large-capacity three-dimensional CT images, then detection accuracy is improved, but computational cost increases

Engineering Contradiction:
Improvelandmark detection accuracyVSAvoidcomputational cost
Core Design Contradiction:
Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The patent applies segmentation by dividing the large-capacity three-dimensional CT image into multiple low-resolution CT images. This allows the CNN model to process smaller image segments instead of the entire large image at once, thereby improving processing speed while maintaining detection accuracy through focused analysis of divided regions.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent extracts only the necessary portions of the CT image by converting three-dimensional CT images into multiple low-resolution CT images that contain relevant landmark information. This extraction approach reduces the computational burden on the CNN model while preserving the essential features needed for accurate landmark detection.

Inventive Principle:
Principle #2Taking out (Extraction)

3Adaptability or versatility

If images from multiple different CT apparatuses are used for learning, then model adaptability is improved, but learning difficulty increases due to different image densities

Engineering Contradiction:
Improvemodel adaptability to different CT apparatusesVSAvoidlearning difficulty
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The patent applies local quality by standardizing the density characteristics of CT images from different apparatuses. Instead of treating all images uniformly, the method adjusts and normalizes the density properties of images from various CT scanners to a consistent standard, enabling the CNN model to learn effectively from diverse sources without being hindered by density variations.

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The patent applies parameter changes by modifying the density parameters of CT images from different apparatuses to achieve uniformity. This involves adjusting image density characteristics through processing techniques that normalize the data, allowing the learning model to handle images from multiple sources with different density profiles effectively.

Inventive Principle:
Principle #35Parameter changes

Data Source

PatentEP4477150B1Artificial intelligence-based head and neck landmark detection method and device
Publication Date: 2026.03.18 3D ONS INC
  • EP4477150B1 patent drawingFigure 1
  • EP4477150B1 patent drawingFigure 2
  • EP4477150B1 patent drawingFigure 3A~3E

AI summary

The present invention relates to a method for detecting a landmark of the head and neck based on artificial intelligence. The method of the present invention includes an operation in which an original computerized tomography (CT) image obtainer obtains an original CT image of the head and neck of a patient, an operation in which an image converter converts the obtained original CT image into a low-resolution CT image, an operation in which a landmark detector detects a first landmark by inputting the low-resolution CT image into a landmark detection model, an operation in which a peripheral area detector detects a peripheral area based on the detected first landmark of the head and neck, an operation in which an image restorer restores an original CT image of the detected peripheral area, and an operation in which the landmark detector detects a second landmark by inputting the restored image of the peripheral area into the landmark detection model. By primarily detecting a landmark by converting a three-dimensional CT image into a low-resolution CT image and secondarily detecting a landmark in relation to a detected specific area, landmarks of the head and neck can be promptly and accurately detected.