Production of a calibration phantom to support diagnosis in computed tomography and analysis method

Abstract

The invention relates to a method for producing a calibration phantom to support diagnosis in computed tomography and an analysis method, which continuously remains inside a CT device, provides calibration in all scans by fully covering a wide spectrum of the HU scale, enables inter-device data standardisation, and has data processing methods that may be developed within this scope, thereby providing different application areas.

Description

[0001] PRODUCTION OF A CALIBRATION PHANTOM TO SUPPORT DIAGNOSIS IN COMPUTED TOMOGRAPHY AND ANALYSIS METHOD

[0002] Technical field:

[0003] The invention relates to a method for producing a calibration phantom to support diagnosis in computed tomography and an analysis method, which continuously remains inside a computed tomography (CT) device, provides calibration in all scans by fully covering a wide spectrum of the Hounsfield Unit (HU scale), enables interdevice data standardisation, and has data processing methods that may be developed within this scope, thereby providing different application areas.

[0004] State of the art:

[0005] In order to increase the accuracy and reliability of Computed Tomography (CT) systems, the development of calibration phantoms is of great importance. Calibration phantoms are standard reference tools used to evaluate and optimise the performance of CT devices. These tools increase the reliability of results and minimise systematic errors by testing the detection and imaging capacities of the devices.

[0006] In the design of calibration phantoms, certain physical and geometric properties are taken into consideration in order for the phantom to be able to simulate all operating conditions of CT devices. In this design process, factors such as material density, geometric shape, and the placement of reference objects contained therein play a critical role. Proper design of the internal structure of the phantom enables consistent results to be obtained under various imaging conditions.

[0007] The effectiveness of developed calibration phantoms is investigated and optimised in accordance with the characteristics of existing CT systems. At this stage, it is necessary that the materials used be selected and positioned in accordance with the energy levels of the CT devices. Proper placement of the reference objects inside the phantom makes it possible to realistically evaluate the performance of the devices.Various analysis methods are used to evaluate the accuracy of calibration phantoms. These methods are applied in order to determine how the phantom is perceived by different CT devices and how reliable the measurement results are. Performance tests and data analysis evaluate the stability of systems over time and increase the effectiveness of calibration procedures.

[0008] In the design of calibration phantoms suitable for clinical needs, international standards and protocols are established. These standards present design and usage criteria for phantoms, thereby ensuring global acceptance of the systems and facilitating the integration of new technologies. In addition, compliance with these standards renders device performance comparable at an international level.

[0009] The selection of materials used in phantom production is considered as a factor that directly affects device performance. The use of durable and high-quality materials ensures that the phantom is long-lasting and reliable. Furthermore, the biocompatibility and non-toxic properties of the materials are of importance for safe use in clinical environments.

[0010] New analysis methods are continuously being developed in order to evaluate the performance of CT systems in greater detail. Modem technologies such as artificial intelligence and machine learning assist in performing calibration processes in a more precise and rapid manner through analyses carried out on large data sets. These innovative methods make it possible to better understand system performance and to carry out effective improvements.

[0011] The effectiveness of calibration phantoms is supported by regular maintenance and updates. Phantoms need to be periodically reviewed and adapted to technological developments. In addition, user training and feedback play a role in increasing the correct use of phantoms and the effectiveness of analysis methods, thereby supporting overall quality and reliability.

[0012] In Computed Tomography (CT) systems, commercial products may be examined in three categories;

[0013] • Traditional quality phantoms included in the first group are not suitable for scanning together with the patient. These phantoms are used to control properties of the CT device such as spatial resolution and Signal / Noise ratio. Inthis manner, information about the condition of the device is provided and calibration of electronic / mechanical components is enabled.

[0014] • The second group consists of tissue / region-specific phantoms that allow scanning together with the patient. They focus on special-purpose measurements such as bone density and are designed to be positioned close to the scanned region. They contain only material representing the tissue targeted for calibration, and calibration is provided only in the vicinity of this material by directly using the shift in the reference material.

[0015] • The phantoms included in the third group are suitable for continuous scanning together with the patient, and within this scope two products have been identified. Both products are formed using only four materials. The software developed to be used together with these phantoms performs direct comparison, matching, and calibration only based on these four values. Apart from this, no advanced method / software support is provided.

[0016] When studies conducted over the last 30 years are examined, deep learning algorithms such as Radiomics and Convolutional Neural Networks (CNNs) have been increasingly used for radiological image classification and outcome prediction; however, one of the fundamental challenges is to establish robustness against changes dependent on the CT device.

[0017] When existing software-based calibration techniques are examined, it has been determined that the most up-to-date approach is the piece-wise linear technique. The results presented as a preliminary study in the project proposal using this technique have demonstrated the benefits of the approach. However, these preliminary studies have also revealed two fundamental points at which the piece-wise linear approach remains limited, and it is anticipated that these limitations reduce calibration performance.

[0018] • The first of these is that, by its nature, the piece-wise linear approach converges linearly between two successive HU values at which the reference material is located. However, this convergence may not always be linear, and in this case a non-linear stretching technique needs to be developed and adapted to the problem. In particular, in phantoms performing sparse sampling in which thereference HU values are far apart from each other, an increase in the margin of error is inevitable.

[0019] • Existing approaches perform point-based calibration by using the mean of the material measurement. However, the standard deviation of the measurement is as critical as the mean value of the material for device calibration and measurement sensitivity. In this case, a method that takes into account the standard deviation in addition to the mean value needs to be developed.

[0020] Although various proposals and applications have been developed in the state of the art for producing a calibration phantom to support diagnosis in computed tomography and for developing an analysis method, these developments are not sufficient. Some applications belonging to inventions developed for this purpose are presented below. In the state of the art, the patent application numbered “CN117442222A” aims to present a new and more suitable method for correcting material-induced effects in computed tomography (CT) images. In particular, this method, which corrects material-induced mass and / or volume effects, calculates the density values of each material by using CT acquisition data of the same object at different energies and determines correction factors in accordance with these calculations. These correction factors assist in the correction of various CT images. This approach eliminates the difficulties generally encountered in calculations carried out by taking into account the effects of various materials, thereby enabling the achievement of more accurate and reliable results. In addition, the invention provides a solution that can be easily integrated into existing devices by enabling such calculations to be performed in a software -based manner in accordance with existing technology.

[0021] In the state of the art, the patent application numbered “AU2012315530A1” describes a system and method aiming to keep radiation doses for Computed Tomography (CT) scanners at the level of “As Low As Reasonably Achievable” (ALARA). The main purpose of the system is to optimise the radiation dose by using the image quality obtained during scanning and the patient’s size data. This method receives the patient’s size data, determines the image quality preferences of a radiologist or a group of radiologists, and generates a target noise equation in accordance with these preferences. The target noise equation is compared with existing noise data, and as a result of this comparison appropriate scanning parameters are calculated. In addition,the system stores the obtained scanning parameters in a database, analyses the results, and continuously monitors scanning quality. This approach ensures the preservation of image quality while minimising the radiation dose, thereby increasing both patient safety and imaging efficiency.

[0022] In the state of the art, the patent application numbered “CN116894878 A” presents a method aiming to correct image artefacts caused by metal objects in real time. In particular, metal objects such as long metal objects and surgical needles cause distortions in computed tomography (CT) images, making it difficult to accurately evaluate anatomical structures. The proposed method identifies artefacts caused by metal objects by using predefined data and performs correction by removing these data from the CT image. This process minimises shadowing and distortions caused by metal objects, thereby enabling clearer and more accurate visualisation of anatomical structures during surgical intervention.

[0023] In the state of the art, there is a need for a method for producing a calibration phantom to support diagnosis in computed tomography and an analysis method that enables a newly acquired image to be selected from an image archive and calibrated by taking as reference an image series from another date in which the same phantom is included, presents the result to the physician through software integrated by creating a transformation between the image series to be selected as reference and the newly acquired scan to be calibrated, includes software that detects different values in the HU values identified in patient images obtained during scanning, is used for the purpose of calibrating CT images obtained at different times, and enables accurate and rapid diagnosis by using an artificial intelligence- and histogram-based calibration algorithm in revealing differences in the images.

[0024] As a result, due to the negative aspects described above and the insufficiency of existing solutions with respect to the subject matter, it has become necessary to carry out a development in the relevant technical field.

[0025] The Aim of the invention:

[0026] The most important aim of the invention is to enable diagnosis to be made rapidly and accurately in computed tomography (CT).Another aim of the invention is to increase measurement sensitivity by presenting to the physician, in addition to the measurements on the calibrated image, the confidence interval of these measurements as a result of taking into account the effect of standard deviation while performing HU value calibration according to reference values, and in this way it is aimed to increase diagnostic performance. Thus, inconsistencies arising from the subjective interpretations of the specialist are reduced, and more accurate and consistent determination of tumour boundaries is enabled.

[0027] Another aim of the invention is that the phantom to be produced consists of samples having a wide range of HU values, that CT images obtained at different times with this phantom are used for calibration purposes, and that an artificial intelligence- and histogram-based calibration algorithm is used in revealing differences in the images, which constitutes the aim of this study.

[0028] Another aim of the invention is to enable diseases to be identified rapidly in CT images obtained by means of the calibration phantom. Thus, the creation of an interface and its presentation for use by the Radiologist are enabled.

[0029] Another aim of the invention is to greatly accelerate the computer-assisted automatic segmentation process and to increase efficiency. Thus, the calibration phantom to be produced, unlike other products present in the sector, continuously remains inside the CT device, provides calibration in all scans by fully covering a wide spectrum of the HU scale, and provides superiority over other products by means of inter-device data standardisation and data processing methods that may be developed within this scope. Another aim of the invention is to enable the creation of improved databases for machine learning techniques while obtaining images with better diagnostic value by means of corrective software in the calculation of CT images that take into account known HU reference values by means of the calibration phantom to be developed in this study. Here, the expression “improved” will enable the elimination of decreases in performance and problems caused in the performance of modem machine learning techniques by data collected from multi-centre and / or different brand / model devices.Description of the figures:

[0030] FIGURE-1 is a drawing providing a top isometric view of the calibration phantom, which will be prepared in the form of a cushion consisting of samples having different HU values, printed by a 3D printer.

[0031] Reference numbers:

[0032] 1: Calibration phantom

[0033] Description of the Invention:

[0034] The invention primarily comprises the development of a phantom in order to be able to perform calibration procedures in a practical manner. In the developed phantom, samples of at least 1 cm3representing the Hounsfield Unit (HU) values of organs and tissues in the body are obtained by using filaments having different properties in a 3D printer. These prepared samples are placed with a certain distribution within a patient cushion, and as a result this developed cushion is used for calibration purposes in our studies.

[0035] In general terms, the invention is the development of a method for producing an imaged calibration phantom to support diagnosis in computed tomography and an analysis method, comprising placement of the calibration phantom, calibration of a newly acquired image, presentation of the result of the image series to be selected as reference and the newly acquired scan to be calibrated to the physician by means of a method to be developed, recording the transformation parameters as a data object in an image archive, and storing them by matching them with the scan to which they are applied.

[0036] Filaments such as LW-PLAfor cylindrical samples having tissue-equivalent HU values and Stonefil filaments for high values such as +1000 HU are used. By means of the LW-PLA filament, a wide range of HU values is obtained at different temperatures and flow rates.

[0037] The method for producing a calibration phantom to support diagnosis in computed tomography comprises the process steps of;• Development of a phantom by the developer in order to be able to perform calibration procedures,

[0038] • Determination of samples of at least 1 cm3representing the Hounsfield Unit (HU) values of organs and tissues in the body in the developed phantom, and • Production of the calibration phantom by preparing the samples for use in the developed studies by placing them with a certain distribution within the patient cushion using filaments having different properties and printing methods in a 3D printer.

[0039] Existing approaches perform point-based calibration by using the mean of the material measurement. However, the standard deviation of the measurement is as critical as the mean value of the material for device calibration and measurement sensitivity. In this case, a method is developed that takes into account the standard deviation in addition to the mean value and, when necessary, additional properties.

[0040] The calibration phantom to support diagnosis in computed tomography and the analysis method comprise the process steps of;

[0041] • Placement of the calibration phantom beneath the anatomical region,

[0042] • Calibration of the newly acquired image with respect to the reference image by using the developed artificial intelligence-supported method,

[0043] • Presentation of the obtained calibrated image to the physician through the software into which it is integrated,

[0044] • Recording the transformation parameters of the calibration process applied to the image as a data object in the image archive, and

[0045] • Storage in the database by matching with the applied scan.

[0046] The calibration of a newly acquired image with respect to a reference image by using the developed artificial intelligence-supported method, which is a process step of the calibration phantom to support diagnosis in computed tomography and the analysis method, comprises the process steps of;

[0047] • Performing scans in which the developed phantom and the standardised phantoms are located together,• After each CT scan, determining the position and boundaries of the reference materials within the calibration phantom and calculating the mean value and standard deviation as Hounsfield value units from within a circle having a diameter of 20 pixels centred at the midpoint of the material,

[0048] • After each CT scan, calculating the mean value and standard deviation as Hounsfield value units from within circles having a diameter of 20 pixels corresponding both to the phantom reference materials and to the values between the reference materials for the regions determined within the standardised phantom,

[0049] • Creating the database by recording the mean value and standard deviation values for each region in each scan by means of repeated scans,

[0050] • Performing supervised (supervised) learning that takes the reference material mean and standard deviation values within the developed phantom as inputs and takes the mean and standard deviation values in the intermediate region of the standardised phantom as outputs when a sufficient amount of data is collected for training the artificial intelligence model, and

[0051] • Performing a new scan including the developed phantom.

[0052] In the process step of performing supervised learning that takes the reference material mean and standard deviation values within the developed phantom as inputs and takes the mean and standard deviation values in the intermediate region of the standardised phantom as outputs when a sufficient amount of data is collected for training the artificial intelligence model, which is a process step of the calibration of a newly acquired image with respect to a reference image by using the developed artificial intelligence-supported method, which is a process step of the calibration phantom to support diagnosis in computed tomography and the analysis method, the input dimension is not limited to the mean and the standard deviation. When required, the dimension of the input and output data can be increased with other properties (energy, homogeneity, entropy). When the trained artificial intelligence model is fed with the presented inputs, it forms a system that produces an interpolation function for each intermediate region. The denser the sampling between the reference materials and the closer the reference materials are to each other in the developed phantom, the higher the success becomes.The process step of performing a new scan including the developed phantom, which is the process step of calibrating a newly acquired image with respect to a reference image by using the developed artificial intelligence-supported method, which is a process step of the calibration phantom to support diagnosis in computed tomography and the analysis method, comprises the process steps of;

[0053] • Determining the reference material boundaries in the developed phantom and calculating the deviation in the mean HU values of the materials,

[0054] • Recalculating the values within the image obtained in the scan by adding them with this deviation value,

[0055] • Calculating the values remaining between the reference material values by means of the artificial intelligence model, and

[0056] • Changing the pixel values corresponding to the calculated values in the image.

[0057] All changes performed are entered into the table known as the Look-Up Table (LUT) in image processing. In this table, one column contains the original pixel values in the image, while the second column contains the calibrated pixel values. The values referred to as transformation parameters are the LUT table.

[0058] The method that is the subject of the invention uses information belonging to data acquired in the past by analysing it by means of artificial intelligence for convergence to the region between the reference materials. In this way, convergence is achieved between two HU values at which the reference materials are located not by a fixed algorithm but by a dynamic approach dependent on data characteristics. Additionally, the convergence of the artificial intelligence model may not be linear, and this particularly increases calibration quality. For this reason, it is ensured that the margin of error is reduced, particularly in phantoms performing sparse sampling in which the reference HU values are far apart from each other.

[0059] The calibration phantom shown in Figure 1 has properties similar to the patient cushion of the CT device and will be produced as a separate cushion containing samples having different HU values, and in CT scans the phantom is placed beneath the relevant anatomical region.A newly acquired image is selected from the image archive and is calibrated by taking as reference an image series from another date in which the same phantom is included.

[0060] The method to be developed presents the result to the physician through the software into which it is integrated by creating a transformation between the image series to be selected as reference and the newly acquired scan to be calibrated.

[0061] The transformation parameters will be recorded in the image archive as a data object compliant with the DICOM Gray Scale Softcopy Presentation State (GSPS) standard and will be stored by matching with the scan to which it is applied.

[0062] When controlled and continuously ongoing data acquisition begins, an algorithm relating to how image calibration values in the region remaining between the measurement points will be determined is prepared. In particular, test cases are created in order to address situations containing shifts in different directions (positive / negative) in different HU regions, and the performance of the algorithm is evaluated. For this purpose, multi-scale radial basis functions that converge to the deviation between the HU values of two reference samples are utilised.

Claims

1. CLAIMS1. A method for producing a calibration phantom to support diagnosis in computed tomography, comprising the process steps of;• development of a phantom by a developer in order to be able to perform calibration procedures,• determination of samples representing the Hounsfield Unit (HU) values of organs and tissues in the body in the developed phantom, and• production of the calibration phantom by preparing the samples using filaments having different properties in a 3D printer.

2. The method for producing a calibration phantom to support diagnosis in computed tomography according to Claim 1 , comprising samples of at least 1 cm3in the process step of determining samples representing the Hounsfield Unit (HU) values of organs and tissues in the body in the developed phantom.

3. The method for producing a calibration phantom to support diagnosis in computed tomography according to Claim 1, wherein the process step of producing the calibration phantom by preparing samples using filaments having different properties in a 3D printer comprises filaments having different properties such as LW-PLA filaments for cylindrical samples and Stonefil filaments for high values such as +1000 HU.

4. The method for producing a calibration phantom to support diagnosis in computed tomography according to Claim 1 , wherein, in the process step of producing the phantom by preparing samples using filaments having different properties in a 3D printer, the samples are placed with a certain distribution within a patient cushion, and the calibration phantom is used in the developed studies as a result.

5. An analysis method of a calibration phantom to support diagnosis in computed tomography, comprising the process steps of;• placement of the calibration phantom beneath the anatomical region,• calibration of the newly acquired image,• presentation of the result to the physician through the artificial intelligence- supported software into which the developed method is integrated, • recording the transformation parameters as a data object in the image archive, and• storage in the database by matching with the applied scan.

6. The analysis method of a calibration phantom to support diagnosis in computed tomography according to Claim 3, wherein the process step of calibrating the newly acquired image with respect to the reference image by using the developed artificial intelligence-supported method comprises the process steps of;• performing scans in which the developed phantom and the standardised phantoms are located together,• after each CT scan, determining the position and boundaries of the reference materials within the calibration phantom and calculating the mean value and standard deviation as Hounsfield value units from within a circle having a diameter of 20 pixels centred at the midpoint of the material,• after each CT scan, calculating the mean value and standard deviation as Hounsfield value units from within circles having a diameter of 20 pixels corresponding both to the phantom reference materials and to the values between the reference materials for the regions determined within the standardised phantom,• creating the database by recording the mean value and standard deviation values for each region in each scan by means of repeated scans, • performing supervised (supervised) learning that takes the reference material mean and standard deviation values within the developed phantom as inputs and takes the mean and standard deviation values in the intermediate region of the standardised phantom as outputs when a sufficient amount of data is collected for training the artificial intelligence model, and• performing a new scan including the developed phantom.

7. The analysis method of a calibration phantom to support diagnosis in computed tomography according to Claim 3 or Claim 4, wherein in the process step of performing supervised learning that takes the reference material mean andstandard deviation values within the developed phantom as inputs and takes the mean and standard deviation values in the intermediate region of the standardised phantom as outputs, when a sufficient amount of data is collected for training the artificial intelligence model, the input dimension is not limited to the mean and the standard deviation, and, when required, the dimension of the input and output data can be increased with other properties such as energy, homogeneity, and entropy.

8. The analysis method of a calibration phantom to support diagnosis in computed tomography according to Claim 3 or Claim 4, wherein the process step of performing a new scan including the developed phantom comprises the process steps of;• determining the reference material boundaries in the developed phantom and calculating the deviation in the mean HU values of the materials, • recalculating the values within the image obtained in the scan by adding them with this deviation value,• calculating the values remaining between the reference material values by means of the artificial intelligence model, and• changing the pixel values corresponding to the calculated values in the image.

Images