Search

KR-20260066657-A - A method for detecting at least one lesion of a patient's pancreas in at least one medical image.

KR20260066657AKR 20260066657 AKR20260066657 AKR 20260066657AKR-20260066657-A

Abstract

The present invention relates to a method for detecting a patient's pancreatic lesion in at least one medical image, implemented by computer means, such as a portal system computed tomography (CT) scan. A segmentation network is trained using a 5-fold cross-validation method. Then, the output of this network is post-processed to extract image features: normalized lesion risk, predicted lesion diameter, and MPD diameters of the pancreatic head, body, and tail. A logistic regression model is calibrated to predict the presence of a lesion based on these features.

Inventors

  • 아비 나데르 클레망
  • 베틸 레베카
  • 로헤 마크 미셸

Assignees

  • 게르브

Dates

Publication Date
20260512
Application Date
20240430
Priority Date
20230502

Claims (15)

  1. A method implemented via computer means to detect at least one lesion of a patient's pancreas in at least one medical image, comprising an inference step including the following steps: (a) predicting at least one first probability map representing the probability that each pixel or voxel of the image is part of a pancreatic lesion, (b) predicting at least one second probability map representing the probability that each pixel or voxel of the image is part of the pancreas, (c) predicting at least one third probability map representing the probability that each pixel or voxel of the image is part of the main pancreatic duct, (d) a step of segmenting the pancreas, main pancreatic duct, and at least one lesion of the image based on the probability map, (e) a step of determining the head, body, and tail portions of the divided pancreas, (f) A step of determining a first feature indicating lesion risk based on a first probability map, (g) A step for determining a second feature indicating the maximum size of the lesion, (h) A step of determining a third feature indicating the maximum size of the main pancreatic duct in the head of the pancreas, (i) A step of determining the fourth feature indicating the maximum size of the main pancreatic duct in the body of the pancreas. (j) A step of determining the fifth feature indicating the maximum size of the main pancreatic duct in the tail portion of the pancreas, (k) a step of determining a first score based on at least one of the first and second features and at least one of the third, fourth and fifth features, wherein the first score represents the probability that the patient has a pancreatic lesion.
  2. In Article 1, The above medical image is a CT scan image.
  3. In Article 1 or Article 2, The above image is a method in which the image is a context CT scan image.
  4. In any one of paragraphs 1 to 3, Step (f) is a method that includes the following sub-steps: - Step to remove all pixels or ploxels outside the segmented pancreas, - For each remaining connection component, a step of calculating the lesion risk by averaging the probabilities of all pixels or voxels and removing connection components with a lesion risk lower than a threshold.
  5. In any one of paragraphs 1 to 4, The second feature is the maximum diameter of the corresponding divided pancreatic lesion.
  6. In any one of paragraphs 1 to 5, A method in which the image is a 3D image, and the maximum diameter of the corresponding segmented pancreatic lesion is determined by measuring the 2D ferret diameter of the segmented lesion along the axial viewpoint for each slice.
  7. In any one of paragraphs 1 to 6, The first, second, and third probability maps are obtained using at least one convolutional neural network or model, e.g., nnUnet.
  8. In any one of paragraphs 1 to 7, A method in which the first, second, and third probability maps are obtained using a plurality or ensemble convolutional neural network or model, each model can predict the first, second, and third probability maps, and the output of the model is averaged in pixel units or voxel units to generate each of the probability maps.
  9. In any one of paragraphs 1 through 8, The above model ensemble is a method trained using a k-fold cross-validation process.
  10. In any one of paragraphs 1 through 9, The image is a 3D image, and step (e) is a method that includes the following sub-steps: - A step of extracting the morphological skeleton of pancreatic divisions slice by slice along the axial viewpoint. - A step of obtaining a 3D skeleton of the pancreas and converting the skeleton into a graph. - Considering the image point located in the far right anterior lower part of the abdomen, the step of identifying the graph point furthest from the abdomen; this point is considered to be the tail tip of the pancreas. - The step of identifying the head of the pancreas as the point on the graph furthest from the tail, - A step to calculate the shortest path between the head and the tail to obtain a centerline that passes through the pancreas and connects the head and the tail ends. - The step of dividing the centerline into head, body, and tail sections. - For each voxel of the pancreatic division, the step of finding the point closest to the centerline and assigning the corresponding head, body, or tail portion.
  11. In any one of Articles 1 to 10, A method in which, for each voxel of a divided main pancreatic duct, the nearest point on the centerline is calculated and the position of the head, body, or tail portion of the divided pancreas is assigned to said voxel according to the position of the nearest point on the centerline, and the diameter of the main pancreatic duct is measured at each of the head, body, and tail portions.
  12. In any one of paragraphs 1 to 11, The method by which the first score is determined using the first logistic regression model.
  13. In any one of paragraphs 1 to 12, A method comprising the step of determining a second score based on at least the third, fourth, and fifth features, wherein the second score represents the probability that the main pancreatic duct will be dilated.
  14. In any one of paragraphs 1 to 13, The second score is determined using the second logistic regression model.
  15. In any one of paragraphs 1 to 14, A method in which the first score or the second score is determined based on at least one other feature from the following list: - Primary statistics of pancreatic diameter, lesion diameter, main pancreatic duct diameter and/or common bile duct diameter, - Location of the lesion, e.g., location of the lesion in the head, trunk, and/or tail portions mentioned above, - At least one radiographic feature of the pancreas, lesion, main pancreatic duct and/or common bile duct, - 3D geometric features of the pancreas, lesions, main pancreatic duct and/or common bile duct, - 2D geometric features of the pancreas, main pancreatic duct, lesions, and/or common bile duct, and/or primary statistics of these features.

Description

A method for detecting at least one lesion of a patient's pancreas in at least one medical image. The present invention relates to a method implemented via computer means to detect at least one lesion of a patient's pancreas in at least one medical image. Pancreatic cancer is currently the 11th most common cancer worldwide and ranks 7th among cancer-related causes of death. With incidence rates projected to increase by up to 78% from 2018 to 2040, pancreatic cancer is emerging as a major medical issue. Combined with the rising incidence rate and a very low 5-year survival rate of 9%, it could become the 3rd leading cause of cancer-related death by 2025. Most patients with early-stage pancreatic cancer present with nonspecific symptoms, and when examination is necessary, they often undergo routine computed tomography (CT) scans during the portal venous phase. In the early stages, image interpretation can be challenging because pancreatic lesions are small (less than 2 cm) and isometric, with reported sensitivities ranging from 58% to 77%. Furthermore, the heavy workload and varying levels of expertise and experience of radiologists can further impact CT scan interpretation. Due to the rapid progression of the disease, pancreatic cancer is mostly detected in late stages where limited treatment options are available, which explains the observed low five-year survival rate. To date, only 10% of patients undergo pancreatectomy, the only available treatment. However, the proportion of patients diagnosed with stage 1A pancreatic cancer has increased over the past few years. Generally, stage 1A pancreatic cancer is characterized by tumors confined to the pancreas, not exceeding 2 cm (0.8 inches) in diameter, and without metastasis to surrounding lymph nodes or other sites. In this stage 1A, patients are more likely to undergo pancreatectomy and adjuvant chemotherapy; consequently, the 5-year survival rate for these patients currently exceeds 80%, highlighting the importance of detecting pancreatic cancer as early as possible. To identify findings that could alert radiologists to the potential presence of pancreatic cancer, the study retrospectively analyzed CT scans of pancreatic cancer patients prior to histopathological diagnosis. There was consensus that subtle secondary signs, such as main pancreatic duct (MPD) dilation, are often observed up to one year after diagnosis. This is because pancreatic cancer is primarily a tubular adenocarcinoma, and the malignant tumor causes MPD stenosis, leading to upstream dilation. Dilation is generally defined as a duct larger than 3 mm in the head of the pancreas and larger than 2 mm in the body and tail. In addition, upstream stenosis, reported as local loss of the MPD lumen, is also a pathological feature. Considering this context, deep learning (DL) methods can play a significant role in the daily practice of radiologists by providing warnings to patients at risk of developing pancreatic cancer. Promising results have been obtained in some pathologies, such as breast cancer, where DL models significantly reduced false positive and false negative rates on two large datasets and greatly reduced the workload of radiologists. These efforts have also been applied to pancreatic cancer, and there have been many attempts to use DL for lesion detection. These studies proposed DL models for the detection of pancreatic neoplasms and validated them on independent patient databases. Although they showed promising results, these approaches relied solely on DL and did not address the need to confirm radiological findings that predict pancreatic cancer, which could improve early lesion detection. Other features, details, and advantages will be illustrated in the following detailed description and drawings: - Figure 1 schematically illustrates an example of a computer device according to the present document. - Figure 2 illustrates the data used to build the training set and the test set. - Figure 3 illustrates the learning step pipeline of the method according to the present document. - Figure 4 illustrates the inference step pipeline of the method according to the present document. - Figure 5 illustrates three different subdivisions of the pancreas. - Figures 6 to 9 illustrate examples of model segmentation, where the left image illustrates the input image and the right image illustrates the segmented elements of the image (red represents the pancreas, green represents the pancreatic lesion, and blue represents the main pancreatic duct). - Figures 10 and 11 illustrate the model performance on the test set. The attached drawings include meaningful colors. Although this application is disclosed in black and white, a color version of the attached drawings has been submitted to the Patent Office. FIG. 1 schematically illustrates an example of a computer device (1) according to the present invention. The computer device (1) includes the following: - Input interface (2), - Memory (3) for stor