Improved lesion detection method

Abstract

The present invention relates to lesion detection. To improve lesion detection, it is proposed to combine deep learning techniques with a contouring strategy, i.e. a subtraction between a contrast-enhanced image and a non-contrast-enhanced image, in order to reflect only the differences between the two images related to the lesion.

Description

Technical Field

[0001] This invention relates to lesion detection, and more specifically, to apparatus and computer-implemented methods for lesion detection, as well as computer program elements and computer-readable media. Background Technology

[0002] Early detection of lesions is crucial for effective treatment. Diagnostic imaging methods such as magnetic resonance (MR), computed tomography (CT), and digital X-ray (DXR) effectively reflect lesions by responding to changes in image contrast compared to normal tissue. For example, multiple sclerosis (MS) is a disease characterized by demyelination of the white matter in the brain within the central nervous system, and white matter lesions exhibit varying contrast on MR images. Due to the extremely high image contrast at the site of white matter lesions, segmentation of the lesions and subsequent analysis are feasible.

[0003] However, image contrast is not always directly discernible due to poor signal-to-noise ratio (SNR) or imaging parameters. The human eye may miss very small lesions when radiologists become fatigued after focusing on reading images. Several contrast-enhancing techniques can be used, such as contrast agents in magnetic resonance imaging (MRI). The use of contrast agents is limited for people with chemical allergies.

[0004] Artificial intelligence (AI) and deep learning have developed rapidly and are now capable of automatically detecting lesions based on high intensity while maintaining image quality. However, detection accuracy decreases as image contrast diminishes. Summary of the Invention

[0005] Improvements in lesion detection may be required. The objectives of this invention are achieved through the subject matter of the independent claims, wherein further embodiments are incorporated in the dependent claims. It should be noted that the aspects described below also apply to the apparatus, computer-implemented methods, computer program elements, and computer-readable media.

[0006] According to a first aspect of the invention, an apparatus for lesion detection is provided. The apparatus includes an input channel configured to receive a first input medical image of a region of interest acquired using a first imaging mode and a second input medical image of a region of interest acquired using a second imaging mode, wherein the second imaging mode has higher sensitivity for lesion detection than the first imaging mode. The apparatus further includes a synthesis module configured to apply a pre-trained deep learning neural network to the first input medical image acquired using the first imaging mode to obtain a second synthesized medical image. The pre-trained deep learning network has been trained based on a training dataset including a historical dataset of medical images of regions of interest acquired using the first imaging mode and medical images of regions of interest acquired using the second imaging mode. The apparatus further includes a contouring module configured to co-register the second input medical image and the second synthesized medical image, and to perform subtraction between the second input medical image and the second synthesized medical image to obtain a subtracted image usable for lesion detection.

[0007] In other words, a method combining deep learning techniques with a contouring strategy (i.e., subtraction between enhanced and unenhanced contrast) is proposed to reflect only the differences between two images relevant to the lesion. For lesions sensitive to a second imaging mode (also denoted as Mode B) but insensitive to a first imaging mode (also denoted as Mode A), deep learning is applied to generate a synthetic image (ImgB_syn) of Mode B from a large dataset. Then, contouring is performed between the actual image of Mode B (ImgB) and the synthetic image of Mode B ({Formula ImgB-ImgB}). This difference reflects the sensitive response of the lesion region in Mode B. This will be described in detail below. Figures 1 to 4 The illustrated embodiments will be explained.

[0008] This subtraction image highlights the lesion area without the need for contrast agent injection. The prominent lesion area in the subtraction image allows for a higher contrast-to-noise ratio (CNR) and can contribute to a more confident and rapid diagnosis.

[0009] According to an embodiment of the present invention, the contouring module is configured to normalize the second input medical image and the second synthesized medical image, and to perform subtraction between the normalized second input medical image and the normalized second synthesized medical image.

[0010] According to an embodiment of the present invention, the device further includes a lesion identification module configured to identify lesions in a subtraction image.

[0011] For example, methods such as classifiers, region growing, neural networks, and deformable models can be used for automated lesion detection.

[0012] According to embodiments of the present invention, the pre-trained deep learning neural network includes at least one of the following: U-Net, and generative adversarial network.

[0013] According to an embodiment of the present invention, the lesion includes multiple sclerosis lesions.

[0014] According to embodiments of the present invention, the first imaging mode and the second imaging mode include at least one of the following:

[0015] - Conventional computed tomography (CT) and spectral CT;

[0016] - Analog positron emission tomography-computed tomography (PET / CT) and digital PET / CT; and

[0017] -T1-weighted magnetic resonance imaging (MRI) and T2-FLAIR MRI.

[0018] For example, conventional CT can sometimes struggle to identify neck lesions due to their location or artifacts from metal and bone. Spectroscopic CT allows the use of multiple spectral results that can enhance lesion visualization, and therefore has higher sensitivity than conventional CT in detecting some lesions.

[0019] For example, digital PET / CT improves the detectability and characteristics of small lesions compared to the same patient acquired on a simulated PET / CT scan. Therefore, digital PET / CT has higher sensitivity for detecting some lesions than simulated PET / CT.

[0020] For example, in T2-FLAIR MRI, MS lesions show high signal intensity in the white matter region, but in T1-weighted MRI, the lesions show only slight signal changes or no signal changes. It should be understood that the proposed device and method can also be used for other types of medical imaging, as long as they can generate images in both modes and the lesion behavior is evident in one mode but not in the other.

[0021] According to an embodiment of the invention, the device further includes an output channel configured to output subtraction images and / or lesion identification results.

[0022] According to an embodiment of the invention, the device further includes a display for displaying subtraction images and / or lesion identification results.

[0023] According to a second aspect of the present invention, a computer-implemented method for lesion detection is provided. The computer-implemented method includes:

[0024] - Receive a first input medical image of a region of interest acquired using a first imaging mode and a second input medical image of a region of interest acquired using a second imaging mode, wherein the second imaging mode has higher sensitivity for detecting lesions than the first imaging mode;

[0025] - A pre-trained deep learning neural network is applied to a first input medical image acquired using a first imaging mode to obtain a second synthetic medical image, wherein the pre-trained deep learning network has been trained based on a training dataset that includes a historical dataset of medical images of regions of interest acquired using the first imaging mode and medical images of regions of interest acquired using the second imaging mode.

[0026] - Jointly register the second input medical image with the second synthesized medical image; and

[0027] - Perform subtraction between the second input medical image and the second synthesized medical image to obtain a subtracted image that can be used for lesion detection.

[0028] According to an embodiment of the present invention, the computer-implemented method further includes the following steps: normalizing the second input medical image and the second synthetic medical image, and subtracting the normalized second input medical image from the normalized second synthetic medical image.

[0029] According to an embodiment of the present invention, the lesion includes multiple sclerosis lesions.

[0030] According to embodiments of the present invention, the first imaging mode and the second imaging mode include at least one of the following:

[0031] - Conventional computed tomography (CT) and spectral CT;

[0032] - Analog positron emission tomography-computed tomography (PET / CT) and digital PET / CT; and

[0033] -T1-weighted magnetic resonance imaging (MRI) and T2-FLAIR MRI.

[0034] According to embodiments of the present invention, the computer-implemented method further includes displaying subtraction images and / or lesion identification results.

[0035] According to a third aspect of the invention, a computer program element is provided for controlling an apparatus according to the first aspect and any related examples, wherein when executed by a processing unit, the computer program element is adapted to perform a method according to the second aspect and any related examples.

[0036] According to a fourth aspect of the present invention, a computer-readable medium storing program elements according to a third aspect of the present invention is provided.

[0037] Advantageously, the benefits provided by any of the above aspects also apply to all other aspects, and vice versa.

[0038] As used herein, the term “image” includes image data, composite image data formed by multiple image data, and other types of data that can be acquired by medical imaging devices such as CT scanners, MRI scanners, etc.

[0039] As used herein, the term "module" may refer to application-specific integrated circuits (ASICs), electronic circuitry, processors (shared, dedicated, or grouped) and / or memories (shared, dedicated, or grouped) that execute one or more software or firmware programs, combinational logic circuitry, and / or other suitable components that provide the described functionality, or a portion thereof.

[0040] It should be understood that all combinations of the foregoing concepts and the additional concepts discussed in more detail below (these are not mutually inconsistent concepts) are considered part of the inventive subject matter disclosed herein. In particular, all combinations of the claimed subject matter appearing in this disclosure are considered part of the inventive subject matter disclosed herein.

[0041] These and other aspects of the invention will become clear with reference to the embodiments described below. Attached Figure Description

[0042] In the accompanying drawings, the same reference numerals often refer to the same parts from different perspectives. Furthermore, the drawings are not necessarily drawn to scale, but rather focus on illustrating the principles of the invention.

[0043] Figure 1 This is a schematic diagram of an apparatus according to some embodiments of the present disclosure.

[0044] Figure 2A and Figure 2B Examples of the first input medical image and the second input medical image are given respectively.

[0045] Figure 3 An example illustrating the training process is provided.

[0046] Figure 4 An example illustrating the subtraction process is provided.

[0047] Figure 5 A flowchart of a method according to some embodiments of the present disclosure is shown. Detailed Implementation

[0048] Figure 1 An apparatus 10 for lesion detection according to some examples of the present disclosure is schematically shown. The apparatus 10 includes an input channel 12, a synthesis module 14, a profiling module 16, and an output channel 18.

[0049] Input channel 12 is configured to receive a first input medical image ImgA (also denoted as Mode A) of the region of interest acquired using a first imaging mode and a second input medical image ImgB (also denoted as Mode B) of the region of interest acquired using a second imaging mode. The second imaging mode (i.e., Mode B) has higher sensitivity for detecting lesions than the first imaging mode (i.e., Mode A). In the example, the first imaging mode is conventional CT, and the second imaging mode is spectral CT. Using spectral data, lesions that are not identified by conventional CT can be seen. In another example, the first imaging mode is analog PET / CT, and the second imaging mode is digital PET / CT. Digital PET / CT can improve the detectability and characteristics of small lesions compared to the same patient acquired on analog PET / CT. In another example, the first imaging mode is T1-weighted MRI, and the second imaging mode is T2-FLAIR MRI.

[0050] Figure 2A and Figure 2B Examples of the first and second input medical images are illustrated separately. Specifically, Figure 2A This illustrates an example of the first input medical image (ImgA) of a patient's brain acquired using T1-weighted MRI. Figure 2B This illustrates an example of a second-input medical image (ImgB) of the brain acquired using T2-FLAIR MRI. (See example...) Figure 2A and Figure 2B As described, the MS lesion showed a high signal in the white matter region in the second input medical image ImgB (i.e., T2-FLAIR_MS image), but the lesion showed only slight signal changes or no signal changes in the first input medical image ImgA (i.e., T1w_MS image), and was therefore not obvious.

[0051] Turn Figure 1 The synthesis module 14 is configured to apply a pre-trained deep learning neural network to a first input medical image ImgA acquired using a first imaging mode (i.e., mode A) to obtain a second synthesized medical image ImgB_syn. The pre-trained deep learning neural network has been trained on a training dataset that includes a historical dataset of medical images of regions of interest acquired using the first imaging mode and medical images of regions of interest acquired using a second imaging mode. Examples of pre-trained deep learning neural networks may include, but are not limited to, U-Net and generative adversarial networks.

[0052] Optionally, such as Figure 1 As described, the apparatus 10 may include a training module 20 for training a deep learning neural network. To train the deep learning neural network, a training image dataset is collected, which includes medical images of regions of interest acquired using a first imaging mode and medical images of regions of interest acquired using a second imaging mode. The training image dataset may include medical images acquired from multiple patients. The medical images in the training image dataset may not have lesions in their regions of interest. This is because the pre-trained deep learning neural network only learns the contrast conversion mapping from mode A to mode B, but does not learn the lesion mapping because the lesions in the mode A image are not obvious.

[0053] Figure 3 An example illustrating the training process is provided. Again, the training process is illustrated using MRI as an example. In this example, the training image dataset includes T1-weighted images as ImgA and T2 FLAIR images as ImgB. For object i in this dataset, there are images of a first imaging mode (i.e., mode A) and images of a second imaging mode (i.e., mode B). For object i in this dataset, the image of the first imaging mode can also be represented by ImgA_i, and the image of the second imaging mode can also be represented by ImgB_i. The training dataset is {ImgA, ImgB} with N objects. A deep learning neural network is then built, trained with ImgA as input and ImgB as the target label, so that for a new object k, ImgA_k can be used to synthesize ImgB_k_syn through the deep learning neural network. Then, the deep learning neural network is built using T1-weighted images as input and T2-FLAIR as the target label.

[0054] The trained deep learning neural network is then applied to the first input medical image ImgA (such as, Figure 3 The T1w_MS image in the image is used to synthesize a second synthetic medical image ImgB_syn (such as, Figure 3 The image shows a T2-FLAIR-MS composite image, but the lesion is not obvious in the image.

[0055] Turn Figure 1 The contouring module 16 is configured to co-register the second input medical image ImgB (with lesions as supersignals) and the second synthetic medical image ImgB_syn (without obvious lesion contrast), and to perform subtraction between the second input medical image ImgB and the second synthetic medical image ImgB_syn to obtain a subtraction image that can be used for lesion detection.

[0056] Figure 4An example illustrating the subtraction process is provided. In this example, the second input medical image ImgB (such as, Figure 4 The actual T2-FLAIR-MS image) includes super-signal contrast in the MS lesion region. On the other hand, the second synthetic medical image ImgB_syn (such as, Figure 4 The synthesized T2FLAIR-MS images only reflect the T2-FLAIR tissue contrast response, but not the contrast differences in MS lesion areas. This is because the trained deep learning neural network only learns the mapping from T1-weighted to T2-FLAIR contrast transformation, and does not learn the lesion mapping due to the inconspicuous appearance in T1-weighted images.

[0057] The contouring module then registers the two images together and performs subtraction. Therefore, MS lesions are highlighted in the subtraction image, while normal tissue areas are eliminated. The subtraction image used for diagnosis is simpler and clearer.

[0058] Optionally, the contouring module 16 can be configured to normalize the second input medical image and the second synthetic medical image, and to perform subtraction between the normalized second input medical image and the normalized second synthetic medical image.

[0059] Optionally, the device 10 may also include a lesion identification module (not shown) configured to identify lesions in the subtraction image. For example, the lesion identification module may use machine learning-based methods for lesion detection. Because subtraction images allow visualization of lesion boundaries, distinguishing them from surrounding tissue, lesion identification can be more accurate, thereby enabling improved pathways for diagnosis, staging, and treatment monitoring.

[0060] Turn Figure 1 Output channel 18 is configured to output the results of subtraction images and / or lesion identification. The device 10 may also include a display (not shown) for displaying the subtraction images and / or lesion identification results.

[0061] Figure 5 A flowchart is shown of a computer-implemented method 100 for lesion detection according to some embodiments of the present disclosure.

[0062] The computer-implemented method 100 can be implemented as a device, module, or related component stored in a set of logic instructions in a non-transitory machine-readable storage medium or computer-readable storage medium such as random access memory (RAM), read-only memory (ROM), programmable ROM (PROM), firmware, flash memory, etc.; implemented as a device, module, or related component in configurable logic such as, for example, programmable logic array (PLA), field-programmable gate array (FPGA), complex programmable logic device (CPLD); implemented as a device, module, or related component in fixed-function hardware logic using circuit technologies such as, for example, application-specific integrated circuit (ASIC), complementary metal-oxide-semiconductor (CMOS), or transistor-transistor logic (TTL) technology; or any combination thereof. Figure 1 Exemplary means for performing the method are described. For example, computer program code for performing the operations shown in method 100 can be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as JAVA, SMALLTALK, C++, Python, and conventional procedural programming languages ​​such as the "C" programming language or similar programming languages.

[0063] In step 110, a first input medical image of the region of interest acquired using a first imaging mode and a second input medical image of the region of interest acquired using a second imaging mode are received. The second imaging mode has higher sensitivity for lesion detection than the first imaging mode. For example, the first imaging mode is conventional CT, while the second imaging mode is spectral CT. For example, the first imaging mode is analog PET / CT, while the second imaging mode is digital PET / CT. For example, the first imaging mode is T1-weighted MRI, while the second imaging mode is T2-FLAIR MRI.

[0064] Figure 2A An exemplary first input medical image is described in the text, while Figure 2B An exemplary second input medical image is illustrated in the diagram.

[0065] In step 120, a pre-trained deep learning neural network is applied to a first input medical image acquired using a first imaging mode to obtain a second synthetic medical image. Examples of pre-trained deep learning neural networks may include, but are not limited to, U-Net and generative adversarial networks.

[0066] The pre-trained deep learning network has been trained on a training dataset that includes a historical dataset of medical images of regions of interest (ROIs) acquired using a first imaging mode and a second imaging mode. The medical images of ROIs acquired using both the first and second imaging modes in the training dataset may include ROIs without lesions.

[0067] Figure 3 An exemplary training process is illustrated.

[0068] In step 130, the second input medical image and the second synthesized medical image are jointly registered.

[0069] In step 140, subtraction is performed between the second input medical image and the second synthesized medical image to obtain a subtracted image that can be used for lesion detection. Optionally, the computer-implemented method 100 may further include the steps of: normalizing the second input medical image and the second synthesized medical image, and performing subtraction between the normalized second input medical image and the normalized second synthesized medical image.

[0070] Figure 4 An exemplary subtraction process is illustrated.

[0071] Optionally, the computer-implemented method 100 may also include the step of identifying lesions in the subtraction image.

[0072] Optionally, subtraction images and / or lesion identification results can be displayed.

[0073] It should be understood that the above operations can be performed in any suitable order, such as sequentially, simultaneously, or a combination thereof, depending on the specific order required, for example, through input / output relationships, where applicable.

[0074] The aforementioned apparatus and method can not only be used to improve the detection of MS disease lesions, but can also be extended to the detection of brain lesions, especially in cases where lesions in MR images do not show typical contrast changes.

[0075] This method can also be extended to other types of medical imaging, as long as it can generate images in two modes, and the lesion behavior is obvious in one mode and not obvious in the other.

[0076] It should also be understood that, unless expressly indicated to the contrary, in any method claimed herein that includes more than one step or action, the order of the steps or actions of the method is not necessarily limited to the order of the steps or actions of the method.

[0077] All qualifiers as defined and used herein should be understood to control dictionary qualifiers, qualifiers in referenced literature, and / or the general meaning of the qualified terms.

[0078] The indefinite articles “a” and “an” used in the specification and claims shall be understood to mean “at least one” unless the contrary is clearly indicated.

[0079] The phrase “and / or” as used herein in the specification and claims should be understood to mean “any one or both” of the elements so combined, that is, elements that exist together in some cases and separately in others. Multiple elements listed with “and / or” should be interpreted in the same way, that is, “one or more” of the elements so combined. Other elements may optionally exist in addition to those specifically indicated by the “and / or” clause, whether related to or unrelated to those specifically indicated.

[0080] As used herein in the specification and claims, “or” should be understood to have the same meaning as “and / or” as defined above. For example, when referring to items in a separate list, “or” or “and / or” should be interpreted as inclusive, meaning that it includes at least one of a plurality of or a list of elements, but also includes more than one and optional additional unlisted items. Only terms that clearly indicate the opposite, such as “only one” or “exactly one”, or when used in the claims, will mean including a plurality of elements or exactly one element from a list of elements. In general, the term “or” as used herein should be interpreted only to indicate an exclusive substitution (i.e., “one or the other but not both”), preceded by an exclusive term such as “any one,” “one of,” “only one of,” or “exactly one of.”

[0081] As used herein in the specification and claims, the phrase "at least one" relating to a list of one or more elements should be understood to mean at least one element selected from any one or more elements in the list, but does not necessarily include at least one element from each element specifically listed in the list, and does not exclude any combination of elements in the list. This limitation also allows for the optional presence of elements other than those specifically identified in the list of elements referred to by the phrase "at least one," whether related to or unrelated to those specifically identified elements.

[0082] In the claims and the foregoing description, all transitional phrases such as "comprising," "including," "carrying," "having," "containing," "involving," "holding," "forming," etc., shall be understood as open-ended, meaning including but not limited to. Only the transitional phrases "constituting" and "substantially constituting" shall be closed or semi-closed transitional phrases, respectively.

[0083] In another exemplary embodiment of the present invention, a computer program or computer program element is provided, characterized in that it is adapted to perform method steps of a method according to an embodiment of the foregoing embodiments on a suitable system.

[0084] Therefore, computer program elements can be stored on a computer unit, which may also be part of an embodiment of the present invention. This computing unit can be adapted to perform or cause the execution of the steps described above. Furthermore, it can be adapted to operate components of the described apparatus. The computing unit can be adapted to automatically operate and / or execute user commands. The computer program can be loaded into the working memory of a data processor. Therefore, a data processor can be configured to execute the methods of the present invention.

[0085] This exemplary embodiment of the invention covers both computer programs that use the invention from the outset and computer programs that convert existing programs into programs that use the invention through updates.

[0086] Furthermore, computer program elements may be able to provide all the necessary steps to implement the exemplary embodiments of the method described above.

[0087] According to another exemplary embodiment of the present invention, a computer-readable medium such as a CD-ROM is provided, wherein the computer-readable medium has computer program elements stored thereon, the computer program elements being described by the preceding portion.

[0088] Computer programs may be stored and / or distributed on suitable media such as optical storage media or solid-state media provided together with or as part of other hardware, but may also be distributed in other forms such as via the Internet or other wired or wireless telecommunications systems.

[0089] However, computer programs can also be provided via networks like the World Wide Web and downloaded from such networks to the working memory of a data processor. According to another exemplary embodiment of the invention, a medium is provided for enabling the downloadability of computer program elements arranged to perform a method according to an embodiment of the foregoing embodiments of the invention.

[0090] While numerous inventive embodiments have been described and illustrated herein, those skilled in the art will readily conceive of a variety of other means and / or structures for performing the functions and / or achieving the results and / or one or more advantages described herein, and each of these variations and / or modifications is considered to be within the scope of the inventive embodiments described herein. More generally, those skilled in the art will readily understand that all parameters, sizes, materials, and constructions described herein are exemplary, and actual parameters, sizes, materials, and / or constructions will depend on the specific application using the teachings of this invention. Those skilled in the art will recognize or be able to determine numerous equivalents of the particular inventive embodiments described herein using only conventional experimentation. Therefore, it should be understood that the foregoing embodiments are presented by way of example only, and embodiments of the invention may be practiced in ways different from those specifically described and claimed within the scope of the appended claims and their equivalents. The inventive embodiments of this disclosure relate to each individual feature, system, article of manufacture, material, tool, and / or method described herein. Furthermore, any combination of two or more such features, systems, articles of manufacture, materials, tools, and / or methods, provided that such features, systems, articles of manufacture, materials, tools, and / or methods do not contradict each other, is included within the inventive scope of this disclosure.

Claims

1. A device (10) for lesion detection, comprising: - Input channel (12) is configured to receive a first input medical image (ImgA) of a region of interest acquired using a first imaging mode and a second input medical image (ImgB) of the region of interest acquired using a second imaging mode, wherein the second imaging mode has higher sensitivity for detecting lesions than the first imaging mode; - A synthesis module (14) is configured to apply a pre-trained deep learning neural network to the first input medical image acquired using the first imaging mode to obtain a second synthetic medical image (ImgB_syn), wherein the pre-trained deep learning network has been trained based on a training dataset, the training dataset including medical images of the region of interest acquired using the first imaging mode and a historical dataset of medical images of the region of interest acquired using the second imaging mode. as well as - Contouring module (16) is configured to co-register the second input medical image and the second synthetic medical image, and to perform subtraction between the second input medical image and the second synthetic medical image to obtain a subtraction image that can be used for lesion detection. The first imaging mode and the second imaging mode include at least one of the following: - Conventional computed tomography (CT) and spectral CT; - Simulated positron emission tomography (PET / CT) and computed tomography (CT) and digital PET / CT; and - T1-weighted magnetic resonance imaging (MRI) and T2-FLAIR MRI.

2. The apparatus according to claim 1, The contouring module is configured to normalize the second input medical image and the second synthetic medical image, and to perform subtraction between the normalized second input medical image and the normalized second synthetic medical image.

3. The apparatus according to claim 1 or 2, further comprising: - A lesion identification module is configured to identify the lesions in the subtraction image.

4. The apparatus according to claim 1 or 2, The pre-trained deep learning neural network mentioned above includes at least one of the following: - U-Net; and - Generative adversarial networks.

5. The apparatus according to claim 1 or 2, The lesions mentioned include lesions from multiple sclerosis.

6. The apparatus according to claim 1 or 2, further comprising: - Output channel (18) is configured to output the subtraction image and / or lesion identification results.

7. The apparatus according to claim 6, further comprising: - A display for showing the subtraction image and / or the lesion identification results.

8. A computer-implemented method (100) for lesion detection, comprising: - Receive (110) a first input medical image of a region of interest acquired using a first imaging mode and a second input medical image of the region of interest acquired using a second imaging mode, wherein the second imaging mode has higher sensitivity for detecting lesions than the first imaging mode; - Apply (120) a pre-trained deep learning neural network to the first input medical image acquired using the first imaging mode to obtain a second synthetic medical image, wherein the pre-trained deep learning network has been trained based on a training dataset, the training dataset including a historical dataset of medical images of the region of interest acquired using the first imaging mode and medical images of the region of interest acquired using the second imaging mode. - Co-register (130) the second input medical image with the second synthesized medical image; as well as - Perform (140) subtraction between the second input medical image and the second synthesized medical image to obtain a subtraction image that can be used for lesion detection. The first imaging mode and the second imaging mode include at least one of the following: - Conventional computed tomography (CT) and spectral CT; - Simulated positron emission tomography (PET / CT) and computed tomography (CT) and digital PET / CT; and - T1-weighted magnetic resonance imaging (MRI) and T2-FLAIR MRI.

9. The computer-implemented method according to claim 8, further comprising: - Normalize the second input medical image and the second synthesized medical image; as well as - Perform the subtraction between the normalized second input medical image and the normalized second synthesized medical image.

10. The computer-implemented method according to claim 8 or 9, The lesions mentioned include lesions from multiple sclerosis.

11. The computer-implemented method according to claim 8 or 9, further comprising: - Display the subtraction image and / or lesion identification results.

12. A computer program product comprising a computer program that, when executed by a processing unit, causes the processing unit to perform the method according to any one of claims 8 to 11.

13. A computer-readable medium storing a computer program that, when executed by a processing unit, causes the processing unit to perform the method according to any one of claims 8 to 11.

Images