Cancer mapping using machine learning

Abstract

A method for mapping cancer includes inputting data elements from a medical image, a biopsy, and a biopsy pathology label into a machine learning model that estimates the probability of clinically significant cancer in a patient, and outputting, via the machine learning model, an estimate of the probability of clinically significant cancer in each voxel of the three-dimensional image.

Description

[Technical Field]

[0001] [CROSS-REFERENCE TO RELATED APPLICATIONS] This application claims priority to U.S. Provisional Patent Application No. 63 / 376,938, filed September 23, 2022, and U.S. Provisional Patent Application No. 63 / 385,757, filed December 1, 2022, the disclosures of each of which are incorporated herein by reference in their entirety.

[0002] FIELD OF THE DISCLOSURE The present disclosure relates generally to systems and methods for detecting cancerous lesions, and more particularly to systems and methods for identifying cancerous lesions using machine learning algorithms. [Background technology]

[0003] Traditionally, the general location of cancer within an organ or other human tissue is identified using medical imaging techniques such as magnetic resonance imaging (MRI) and then treated using surgery, radiation therapy, chemotherapy, hormone therapy, and / or other methods. Many of these therapeutic therapies can harm healthy tissue in the organ or the healthy tissue surrounding the cancerous tissue, thereby causing irreparable and unnecessary damage if introduced into areas other than the cancerous region. It can be difficult for physicians to accurately predict the extent of cancerous tissue within an area, including determining the exact boundary between the cancerous cells and healthy tissue. Often, physicians must overestimate the size of the cancerous lesion to ensure all cancer cells are removed, which can result in the detriment of removing some surrounding healthy tissue as well. However, depending on the size of the cancerous lesion and the size of the organ in which the cancer resides, this removal of healthy tissue can be detrimental to the continued function of the organ.

[0004] One example of such an organ is the prostate, which is an oval-shaped gland in men that produces semen. Prostate cancer is the most common cancer in men after skin cancer. Prostate cancer can be detected in local or regional stages, representing stages I, II, and III. Due to the location and anatomical structure of the prostate, treatment of cancerous lesions on or within the prostate is difficult and often results in damage to the prostate area or adjacent organs other than the cancerous lesion.

[0005] For this and other reasons, improvements are needed in the area of ​​detection and characterization of cancerous lesions on and within the prostate gland. Summary of the Invention

[0006] In at least one example of the present disclosure, a device for mapping cancer may include a processor electrically coupled to a memory component that stores electronic instructions that, when executed by the processor, cause the device to execute a machine learning algorithm configured to receive input and generate output based on the input, the input including data elements from a medical image, and the output including an estimate of the probability of clinically significant cancer in each voxel of a three-dimensional image.

[0007] In one example, the input further includes prostate specific antigen (PSA), and the probability of clinically significant cancer includes a probability of clinically significant prostate cancer (csPCa). In one example, the output further includes a cancer estimation map (CEM). In one example, the CEM shows a color-coded heat map representing the probability of cancer at each voxel of the three-dimensional image. In one example, the medical image is an MRI image of a patient's anatomy. In one example, the anatomical structure includes a prostate. In one example, the CEM includes a lesion outline representing a lesion size of a cancerous lesion shown in the three-dimensional image. In one example, the output further includes a visual curve representing the encapsulation confidence score versus the lesion size. In one example, the visual curve includes points representing a particular lesion size and a particular encapsulation confidence score. In one example, the points are configured to be visually manipulated along the visual curve to change the particular lesion size and the particular encapsulation confidence score represented by the points, and manipulation of the points alters the lesion contour.

[0008] In at least one example of the present disclosure, a method for mapping cancer includes inputting data elements from a medical image into a machine learning model that estimates the probability of clinically significant cancer in a patient, and generating, via the machine learning model, an output including an estimate of the probability of clinically significant cancer in each voxel of a three-dimensional image.

[0009] In one example, the method further includes inputting prostate-specific antigen (PSA) data elements from a biopsy and a biopsy pathology label into the machine learning model. In one example, the machine learning model is trained based on a population dataset including the data elements. In one example, the output includes a visual representation of the three-dimensional image having a color-coded heat map representing the probability of clinically significant cancer at each voxel.

[0010] In at least one example of the present disclosure, a method for mapping cancer includes inputting data elements from a medical image into a machine learning model that estimates the probability of clinically significant cancer in a patient; and displaying a visual representation of the estimated probability of clinically significant cancer at each voxel of a three-dimensional image. The visual representation may include a cancer estimation map (CEM) showing a color-coded heat map representing the probability of clinically significant cancerous lesions overlaid on the image, the CEM including a lesion contour representing the size of a cancerous lesion and a curve representing an encapsulation confidence score versus the size, the curve including points representing the lesion size and the encapsulation confidence score. In such an example, the points are configured to be visually manipulated along the curve to change the lesion size and the encapsulation confidence score represented by the points, and manipulation of the points alters the lesion contour.

[0011] In one example, the method further includes displaying an interventional instrument at a position relative to the image. In one example, the position of the interventional instrument is configured to be changed relative to the image. In one example, the method further includes displaying a position of a biopsy core overlaid on the image. [Brief explanation of the drawings]

[0012] The present disclosure will be readily understood by the following detailed description taken in conjunction with the accompanying drawings, in which like reference numerals designate like structural elements and in which:

[0013] [Figure 1] We present a framework for a machine learning model that estimates the probability of clinically significant cancer.

[0014] [Figure 2] 1 illustrates exemplary inputs and outputs of a machine learning model that estimates the probability of clinically significant cancer.

[0015] [Figure 3]Indicates an MRI data element.

[0016] [Figure 4A] As output of the machine learning model, additional data elements are shown, including prostate segmentation and region of interest (ROI). [Figure 4B] As output of the machine learning model, additional data elements are shown, including prostate segmentation and region of interest (ROI).

[0017] [Figure 5A] 1 shows csPCa-positive and csPCa-negative biopsy core data elements. [Figure 5B] 1 shows csPCa-positive and csPCa-negative biopsy core data elements.

[0018] [Figure 6] 1 shows biopsy data elements from a biopsy location merged with MRI data elements.

[0019] [Figure 7] Cancer estimation maps (CEM) are shown.

[0020] [Figure 8] 1 shows a patient-specific chart graphic user interface.

[0021] [Figure 9A] The lesion is shown with the CEM and various lesion contours selected from various points on the curve of encapsulation confidence score versus lesion size. [Figure 9B] The lesion is shown with the CEM and various lesion contours selected from various points on the curve of encapsulation confidence score versus lesion size. [Figure 10A] The lesion is shown with the CEM and various lesion contours selected from various points on the curve of encapsulation confidence score versus lesion size. [Figure 10B]The lesion is shown with the CEM and various lesion contours selected from various points on the curve of encapsulation confidence score versus lesion size.

[0022] [Figure 11] 3D reconstructed pathological areas are shown.

[0023] [Figure 12] Illustrates the selection of interventional devices.

[0024] [Figure 13] 1 shows the placement of interventional tools in a 3D representation of the prostate segmentation. [Figure 14] 1 shows the placement of interventional tools in a 3D representation of the prostate segmentation. [Figure 15] 1 shows the placement of interventional tools in a 3D representation of the prostate segmentation. [Figure 16A] 1 shows the placement of interventional tools in a 3D representation of the prostate segmentation. [Figure 16B] 1 shows the placement of interventional tools in a 3D representation of the prostate segmentation.

[0025] [Figure 17] 1 shows an exemplary "half-lobe" boundary.

[0026] [Figure 18] Illustrates isotropic expansion techniques.

[0027] [Figure 19] 1 shows a visual representation and implementation of software and a user interface for mapping cancer using machine learning algorithms. [Figure 20] 1 shows a visual representation and implementation of software and a user interface for mapping cancer using machine learning algorithms. [Figure 21]1 shows a visual representation and implementation of software and a user interface for mapping cancer using machine learning algorithms. [Figure 22] 1 shows a visual representation and implementation of software and a user interface for mapping cancer using machine learning algorithms. [Figure 23] 1 shows a visual representation and implementation of software and a user interface for mapping cancer using machine learning algorithms. [Figure 24] 1 shows a visual representation and implementation of software and a user interface for mapping cancer using machine learning algorithms. [Figure 25] 1 shows a visual representation and implementation of software and a user interface for mapping cancer using machine learning algorithms. [Figure 26] 1 shows a visual representation and implementation of software and a user interface for mapping cancer using machine learning algorithms. [Figure 27] 1 shows a visual representation and implementation of software and a user interface for mapping cancer using machine learning algorithms. [Figure 28] 1 shows a visual representation and implementation of software and a user interface for mapping cancer using machine learning algorithms. DETAILED DESCRIPTION OF THE INVENTION

[0028] Reference will now be made in detail to exemplary embodiments, as illustrated in the accompanying drawings. It should be understood that the following description is not intended to limit the embodiments to any single preferred embodiment. On the contrary, it is intended to cover alternatives, modifications, and equivalents as may be included within the spirit and scope of the described embodiments, as defined by the appended claims.

[0029] The following disclosure relates generally to systems and methods for detecting and characterizing cancerous lesions. More particularly, the disclosure relates to systems and methods for identifying and characterizing cancerous lesions using machine learning algorithms.

[0030] Conventional MRI and biopsy techniques used to detect cancerous lesions provide a rough localization of the cancerous lesion, indicating its presence. After cancer is detected, radiation therapy, chemotherapy, hormonal therapy, surgery, resection therapy, and / or other methods of cancer treatment are applied. This treatment approach can lead to unnecessary exposure of tissues or organs with cancerous lesions to excessive radiation or chemotherapy, or unnecessary removal of healthy tissue surrounding the cancerous cells.

[0031] While the methods and systems described herein for detecting and mapping cancer in a patient may be applied to many or all forms of cancer, one example is the detection and mapping of prostate cancer. Traditional methods may lead to overexposure of the prostate to radiation or ablative therapy, or removal of healthy prostate gland, which may adversely affect urinary, sexual, and / or bowel function, thereby causing a decrease in the quality of life of affected patients.

[0032] The methods and systems for detecting and mapping cancer described herein can more accurately identify cancerous lesions and minimize adverse effects on surrounding healthy tissue, including methods for increasing the reliability of mapped thresholds for lesions. In one example of the methods described herein, a machine learning algorithm can receive MRI and biopsy data including biopsy pathology labels, along with associated encapsulation confidence scores, as inputs for determining lesion thresholds and cancer encapsulation probabilities, minimizing the risk of overexposure and overresection during resection, radiation, and / or surgical intervention. The machine learning algorithms described herein can be trained based on large population datasets to increase the accuracy of cancer mapping, confidence scores, and threshold boundaries. In at least one example, the algorithm can output an estimate of the probability of clinically significant cancer in each voxel of a three-dimensional image.

[0033] These lesion encapsulation boundaries and associated confidence scores can be visually presented to a physician to convey important information used to determine optimal intervention strategies. These visual outputs can include a cancer estimation map (CEM), which shows a heat map of lesion locations overlaid on a medical image of the patient's anatomy, and a three-dimensional cancer lesion contour (CLC), which surrounds areas of elevated cancer probability. By mapping cancer as described herein, the CEM can indicate the spatial probability of tumor presence, while the CLC can indicate the estimated tumor size. While the example system described herein includes CEM, the system described herein can be applied to both CEM mapping and / or CLC generation. Additionally, the system described herein can visually output a plot or curve of encapsulation confidence versus lesion size. These outputs can be modified by the physician as deemed appropriate during analysis when balancing the risk of removing or affecting healthy tissue versus the risk of missing cancerous cells during treatment, as presented by the confidence score curve.

[0034] In at least one example described herein, CEM can be used, alone or in combination with other factors, to assess the stage of a cancer. In at least one example, the cancer is prostate cancer and the stage assessment includes an estimation of the probability and / or location of extraprostatic extension. In at least one example, CEM can be used, alone or in combination with other factors, to assess a patient's suitability for a course of treatment.

[0035] In some examples, inputs from at least one, two, or more data elements, such as medical images including MRI images, X-rays, and ultrasound images, other related medical images, follow-up biopsies, biopsy pathology, biopsy core locations, fusion biopsy data, biomarkers such as PSA, patient demographics such as age, genomic markers, and / or other inputs, can be utilized by a machine learning algorithm that can output an estimate of clinically significant cancer in each voxel of the three-dimensional image to generate the CEM, CLC, and encapsulation confidence scores described above. The estimate of clinically significant cancer can be used to identify and target therapeutic therapies (e.g., chemotherapy, radiation, surgery, etc.). In one example, the encapsulation confidence score represents an estimated probability that the lesion contour encompasses all csPCa.

[0036] The machine learning algorithm of the present disclosure can be trained based on a large population dataset to refine an individual CEM for a particular patient. Thresholding of the CEM can be based on this population analysis. This training dataset can be used as a ground truth for training the algorithm and can include some or all of the inputs listed above for a large population, as well as other inputs specific to a particular type of cancer or other inputs including post-care data and outcomes. The algorithm can then distinguish the probability of cancer at any point within a particular anatomical structure of the patient. In addition, a smaller sub-dataset using surgical data can be used as a "fine-tuning" dataset for the algorithm, allowing for the estimation of tumor encapsulation probability for a particular CLC and patient.

[0037] While the described methods and systems may be applied to many or all cancerous lesions, prostate cancer is described herein as an example for purposes of explanation and illustration. The machine learning algorithms described herein may include additional inputs beyond those described above, for example, inputs specific to a particular type of cancer. In the case of prostate cancer, the above inputs may be combined with additional data inputs, such as prostate specific antigen (PSA) levels, and processed by the machine learning algorithm to provide specific areas within the prostate that are affected, which may lead to more effective treatment and reduce the impact of urinary, sexual, and / or bowel complications.

[0038] In another example, inputs from one, two, or more data elements are analyzed to generate an output that includes a cancer estimation map (CEM) (alternatively referred to as a cancer probability map (CPM)). Within the context of this disclosure, it should be understood that the terms cancer estimation map (CEM) and cancer probability map (CPM) are used interchangeably and define the probability of clinically significant prostate cancer in each voxel of a three-dimensional image.

[0039] In one example, the method further includes displaying a cancer estimation map, an encapsulation confidence score, or metadata and prediction statistics derived from the third-order statistical model. In one example, the metadata includes an encapsulation confidence score representing the probability of all clinically significant cancer contained within a specified lesion contour. In one example, the metadata includes the volume of the lesion contour. In one example, the prediction statistics include an estimate of tumor volume. In one example, the prediction statistics include an estimate of the stage of the cancer. In one example, the prediction statistics include an estimated probability of extracapsular extension, with or without the expected location of the extracapsular extension. In one example, the prediction statistics include an estimate of the patient's suitability for a course of treatment, such as ablative therapy, radiation, radical prostatectomy, or active surveillance. In one example, the prediction statistics include an estimated outcome of a course of treatment, such as the need for additional treatment, the probability of biochemical recurrence or metastasis, the probability of treatment-related side effects, or the probability of death.

[0040] In another example, the output of the machine learning algorithm is further processed, or a second machine learning algorithm is used, to convey additional metadata or information beyond the visual representation of the CEM, CLC, and ECS. For example, an automatically estimated suitability of a patient for a particular treatment (radiation, surgery, ablative therapy, etc.) may be displayed, with or without an estimate of the probability of success for that treatment. In another example, an estimated cancer stage may be displayed, with or without localization and quantification of potential sites of invasive cancer on or beyond the organ of interest. Such information may be used to further help identify and narrow treatment options (e.g., chemotherapy, radiation, surgery, etc.).

[0041] These and other embodiments are discussed below with reference to Figures 1-28. However, those skilled in the art will readily appreciate that the detailed description provided herein with respect to these figures is for illustrative purposes only and should not be construed as limiting. Furthermore, as used herein, a system, method, article, component, feature, or sub-feature that includes at least one of a first option, a second option, or a third option should be understood to refer to a system, method, article, component, feature, or sub-feature that may include one of each listed option (e.g., only one of the first options, only one of the second options, or only one of the third options), multiple of a single listed option (e.g., two or more of the first options), two options simultaneously (e.g., one of the first options and one of the second options), or a combination thereof (e.g., two of the first options and one of the second options).

[0042] While the embodiments described herein are susceptible to various modifications and alternative forms, specific embodiments have been shown by way of example in the drawings and are herein described in detail. However, the exemplary embodiments described herein are not intended to be limited to the particular forms disclosed. Rather, the present disclosure covers all modifications, equivalents, and alternatives falling within the scope of the appended claims.

[0043] 1 illustrates a high-level block diagram of a computer system 100 that may be used to implement embodiments of the present disclosure. In various embodiments, computer system 100 may include different sets and subsets of the components illustrated in FIG. 1. As such, FIG. 1 illustrates various components that may be included in various combinations and subsets based on the operations and functionality performed by system 100 in different embodiments. It should be noted that when described or referenced herein, the use of articles such as "a" or "an" is not to be construed as limiting to only one, but instead is intended to mean one or more, unless specifically stated otherwise herein.

[0044] Computer system 100 may include a central processing unit (CPU) or processor 102, a power supply 108, an electronic storage device 110, a network interface 112, an input device adapter 116, and an output device adapter 120 connected for electrical communication via a bus 104 to a memory device 106. For example, one or more of these components may be connected to each other via a board (e.g., a printed circuit board or other substrate) that supports bus 104 and other electrical connectors that provide electrical communication between the components. Bus 104 may include a communication mechanism for communicating information between portions of system 100.

[0045] The processor 102 may be a microprocessor or similar device configured to receive and execute a set of instructions 124 stored by the memory 106. The memory 106 may be referred to as a main memory, such as a random access memory (RAM), or another dynamic electronic storage device for storing information and instructions to be executed by the processor 102. The memory 106 may also be used to store temporary variables or other intermediate information during execution of instructions to be executed by the processor 102. The power source 108 may include a power supply capable of providing power to the processor 102 and other components connected to the bus 104, such as a connection to an electrical power grid or a battery system.

[0046] Storage device 110 may include read-only memory (ROM) or another type of static storage device coupled to bus 104 for storing static or long-term (i.e., non-dynamic) information and instructions for processor 102. For example, storage device 110 may include a magnetic or optical disk (e.g., a hard disk drive (HDD)), solid-state memory (e.g., a solid-state disk (SSD)), or similar device. Instructions 124 may include information for performing processes and methods using components of system 100.

[0047] Network interface 112 may include an adapter for connecting system 100 to external devices via a wired or wireless connection. For example, network interface 112 may provide connectivity to a computer network, such as a cellular network, the Internet, a local area network (LAN), a separate device capable of wirelessly communicating with network interface 112, other external devices or network locations, and combinations thereof. In one exemplary embodiment, network interface 112 is a wireless networking adapter configured to connect via Wi-Fi, Bluetooth, Bluetooth mesh, Bluetooth mesh, or related wireless communication protocols to another device capable of interfacing using the same protocol. In some embodiments, a network device or set of network devices in network 126 may be considered part of system 100. In some instances, a network device may be considered connected to, but not a part of, system 100.

[0048] The input device adapter 116 may be configured to provide connections to various input devices for the system 100, such as, for example, a touch input device 113 (e.g., a display or display assembly), a keyboard 114 or other peripheral input device, one or more sensors 128, associated devices, and combinations thereof. In some configurations, the input device adapter 116 may include a touch controller or similar interface controller as described above. The sensor 128 may be used to detect physical phenomena (e.g., light, sound waves, electric fields, forces, vibrations, etc.) in the vicinity of the computing system 100 and convert those decreases into electrical signals. The keyboard 114 or another input device (e.g., a button or switch) may be used to provide user input, such as input regarding settings for the system 100.

[0049] Output device adapter 120 may be configured to provide system 100 with the ability to output information to a user, for example, by providing visual output using one or more displays 132, by providing audible output using one or more speakers 135, or by providing touch-detected haptic feedback via one or more haptic feedback devices 137. Other output devices may also be used. Processor 102 may be configured to control output device adapter 120 to provide information to a user via an output device connected to adapter 120.

[0050] Any of the features, components, and / or portions shown in Figure 1, including their arrangement and configuration, either alone or in any combination, may be included in any of the other example devices, features, components, and portions shown in other figures. For example, computing system 100 may be used to run the algorithms described herein and display the visual representations described herein and shown in other figures. Similarly, any of the features, components, and / or portions shown in other figures, including their arrangement and configuration, either alone or in any combination, may be included in the example devices, features, components, and portions shown in Figure 1.

[0051] FIG. 2 illustrates an exemplary data flow diagram 200 for a machine learning model for estimating the probability of clinically significant cancer. By way of example, the systems described herein are described with reference to prostate cancer. However, the systems and methods described herein may be applied to other types of cancer. In at least one example, a software program may utilize as input 202 at least one, two, or more data elements, including an MRI data element 204, a biopsy pathology data element 206, a prostate-specific antigen (PSA) data element 208, and / or data elements for determining the probability of cancer. The use of prostate cancer detection, and specifically PSA, as an input to the systems described herein is exemplary only and not meant to be limiting. Rather, as noted above, the systems and methods described herein may be applied to other types of cancer, and other types of antigens, imaging modalities, genetic information, demographic data, or biomarkers indicative of other types of cancer may be used as input to the system. In one example, input 202 may exclude PSA data element 208.

[0052] In at least one example, the data elements 204, 206, 208 may serve as input 202 to a machine learning model 210. The machine learning model may be a single model, multiple models operating in series or in parallel, and / or one or more models with additional post-processing analysis. In one example, the machine learning model 210 may then estimate the probability of clinically significant cancer. In one example, clinically significant cancer may be defined as Gleason score group 2 or higher in the case of prostate cancer. In at least one example, the machine learning model 210 may then provide an output 216 including an estimate of the probability of clinically significant cancer at each voxel of the 3D image, defined herein as a cancer estimation map (CEM) 212. The final output may be a lesion contour 214. The lesion contour may be a 3D surface generated by thresholding the cancer probability. The output may include additional data related to or derived from the machine learning model, such as an estimated probability of tumor encapsulation, an estimate of tumor stage, an estimate of the patient's suitability for a particular therapy (surgery, radiation, ablative therapy, etc.), tumor segmentation, and / or segmentation of anatomical structures (prostate, urethra, bladder, seminal vesicles, prostatic zone, vas deferens, rectum, pelvis, etc.).

[0053] Any of the features, components, and / or portions shown in Figure 2, including their arrangement and configuration, either alone or in any combination, may be included in any of the other example devices, features, components, and portions shown in other figures. Similarly, any of the features, components, and / or portions shown in other figures, including their arrangement and configuration, either alone or in any combination, may be included in the example devices, features, components, and portions shown in Figure 2.

[0054] 3-28 illustrate examples of the output 216 shown in FIG. 2 of the machine learning algorithm, as well as additional data related to or derived from the machine learning model. The outputs shown in FIGS. 3-28 may be visually represented to a physician or other user on a display screen when a software program executing the machine learning algorithm or the output of the machine learning algorithm is run by a computing device. In one example, as shown in FIG. 3, the input 302 may include a magnetic resonance image (MRI) 300 of a patient, which is utilized as an MRI data element 304 that may be displayed. The MRI data element 204 may include information related to the MRI coordinate space and may be one of the inputs 202 to the machine learning model 210 as discussed in FIG. 2. The MRI data element 204 may be one or more MRI sequences (T2-weighted, diffusion-weighted, perfusion-weighted, etc.) derived from the same patient.

[0055] Figure 4A shows an example prostate MRI 300, as previously discussed in Figure 3, including an example prostate segmentation 402 identified by the machine learning model 210. Figure 4B shows an example prostate MRI 300 of a patient, including an example prostate segmentation 402 and a region of interest (ROI) 404 identified by the machine learning model 210 as output.

[0056] FIG. 5A illustrates a csPCa-positive biopsy core 502 derived from a biopsy system, e.g., a biopsy system that registers or fuses biopsy location with MRI data elements. For example, a patient may receive a biopsy from a biopsy system with an identified csPCa-positive core, and the core location on the prostate may be digitally transcribed onto an MRI image relative to the prostate segment 402, as discussed in FIG. 4A. Similarly, FIG. 5B illustrates a csPCa-negative biopsy core 504 identified by a biopsy system that can track the core's location within the prostate. The location data for the csPCa-negative biopsy core 504 is received as input to the machine learning model 210 and is represented in a different color than the csPCa-positive biopsy core 502. For example, the csPCa-positive biopsy core 502 may be represented in red, while the csPCa-negative biopsy core 504 may be represented in blue, thereby creating a visually distinct distinction between the positive and negative biopsy cores 502, 504. Additional cores that fit into other categories may be represented in one or more tertiary colors. For example, biopsy cores containing clinically insignificant cancer may be represented in orange.

[0057] As shown in FIG. 6, each biopsy core 502, 504 is labeled with one, two, or more attributes 602, such as Gleason score, percent cancer, length of cancer, and length of core, determined and recorded in the pathology report by a pathologist or a pathology analysis algorithm.

[0058] 7 provides a visual representation of a cancer estimation map (CEM) 702 as one exemplary output of the machine learning model 210. The CEM 702 identifies the probability of cancerous locations based on a color gradient (e.g., a heat map) that provides a visual representation of the probability of cancerous tissue at a particular location relative to the prostate segment 402. For example, as discussed in FIG. 2, the machine learning model 201 may receive one, two, or more inputs (e.g., follow-up biopsies, biopsy pathology, and prostate-specific antigen (PSA)) and estimate the prostate segmentation 402, region of interest 404, and / or cancer probability map 702 as output 216. The outputs 402, 404, 702 may increase the accuracy and effectiveness of prostate cancer treatment and / or reduce the amount of prostate tissue removed or damaged when compared to conventional prostate treatment methods.

[0059] The clinician is also presented with a patient-specific chart 800 (alternatively referred to as a Marks reliability curve) depicting an encapsulation confidence curve 806, as shown in FIG. 8 . The x-axis 802 of the patient-specific chart 800 represents the percent of prostate voxels encapsulated by iterative thresholding of the CEM 702. The y-axis 804 of the patient-specific chart 800 represents the encapsulation confidence score, i.e., the confidence that all cancerous cells in the lesion will be encapsulated using each CEM threshold, which ranges from zero to 100 and is expressed as a percentage. The encapsulation confidence score is based on a lookup table correlating the probability of csPCa encapsulation against the CEM threshold. The lookup table provides the clinician with data derived from a retrospective study of pathology data of whole tissue specimens.

[0060] 9A, the machine learning model 210 generates default points 902 in a patient-specific chart 800. A default lesion contour is selected that maximizes the encapsulation confidence score 804, represented on the y-axis, while minimizing the lesion size 802, represented on the x-axis.

[0061] The lesion contour shown in Figure 9B is generated after a point on the patient-specific chart is selected. For example, user-selected point 904 on patient-specific chart 800 selects a smaller lesion size 802 (shown in Figure 8), represented on the x-axis, which reduces the encapsulation confidence score 804 (also shown in Figure 8), represented on the y-axis. User-selected point 904 results in a lower encapsulation confidence than default point 902. The resulting lesion contour size, represented by lesion contour 906, shown in Figure 9B, is overlaid on CEM 702 to create a visual representation of the selected lesion contour size relative to CEM 702, which effectively allows the clinician to "fine-tune" the lesion contour size represented by lesion contour 906, increasing the probability of successful treatment or reduced treatment volume.

[0062] Similarly, Figures 10A-10B show a user-selected point 904 having a greater encapsulation confidence score 804 than the default point 902, thereby increasing the lesion contour size on the X-axis, represented by the lesion contour 906 on the CEM 702 above the default point 902 shown on the curve.

[0063] The user freely adjusts the lesion contour 906, which represents the lesion contour size in the CEM 702, by selecting any point on the encapsulation confidence curve 806. The encapsulation confidence curve 806 promotes a balance between the probability of csPCa encapsulation versus the lesion contour size. For example, as shown in FIGS. 9A-9B, the user may scale down the lesion contour size by selecting a lower CEM 702 threshold and subsequently update the encapsulation confidence score 804. Conversely, as shown in FIGS. 10A-10B, the user may scale up the lesion contour size 96 by making a selection that increases the encapsulation confidence score 804. The lesion contour size selected by the user may depend on the patient's anatomy, the physician's expertise, the type of intervention planned, and other factors.

[0064] As shown in Figure 11, the cancer probability map 702 and lesion contours are evaluated using pathology of the whole tissue specimen. The MRI-registered, 3D-reconstructed pathological tumor region is used to define a ground truth csPCa, which allows for accurate and objective evaluation of key software features such as the encapsulation confidence curve 806.

[0065] As shown in FIG. 12 , once the cancerous lesion outline size is identified, the user can place a virtual interventional device with customizable dimensions. In one example, the device can represent an interstitial catheter, and the intervention can be thermal ablation of the cancerous tissue. The user can select an instrument that can apply the desired ablation size. For example, if a smaller cancerous lesion outline size is identified, a smaller ablation size can be applied. If a larger cancerous lesion outline size is identified, a larger ablation size can be applied. The ablation size location and orientation are user-selectable, and the placement of the interventional device is customizable, allowing the user to choose manual or semi-automatic placement of the interventional device.

[0066] As shown in FIGS. 13-16B, a user may review a prostate segmentation 1302. In another example, a user may review the prostate segmentation 1302, a cancerous lesion contour 1305, and an interventional device 1304. The location and volume of the interventional device 1304 are identified and digitally represented to the user. In one example, the interventional device 1304 represents a probe for causing tissue ablation, and the ablation volume associated with each probe is displayed against the image. In one example, the location and / or ablation volume of the interventional device 1304 is compared to the locations of other anatomical structures. In one example, the configuration of the interventional device 1304 and other anatomical structures may be identified as potential sources of safety or efficacy concerns.

[0067] In particular, Figures 16A and 16B show a CEM 1301 and a corresponding medical image 1307 showing a prostate segmentation 1302 and a virtual interventional device 1304. Figure 16A shows a top view of the medical image 1307, and Figure 16B shows a side view thereof, showing three-dimensional information and properties of the displayed segmentation 1302 and interventional device 1304. The medical image 1307 and the CEM 1301 may be shown side-by-side to provide visual context to an expert. The prostate segmentation 1302 may be represented in three-dimensional space, and the interventional device 1304 may be placed within the virtual three-dimensional space of the segmentation 1302. The position of the interventional device 1304 overlaid on or within the three-dimensional segmentation 1302 may correspond to the recommended position of the actual instrument used during the intervention. The CEM 1301 may include a segmentation 1302, a lesion contour 1305, and a region of interest 1309, as described elsewhere herein.

[0068] Figure 17 shows the most conventional approach to prostate cancer treatment, defined as "hemi-lobe" boundaries. Studies have shown that nearly half of cases of unilateral cancer actually have bilateral cancer. In this example, both hemi-lobe boundaries in the right or frontal hemisphere are dysfunctional, demonstrating the need for a more comprehensive and patient-specific approach to cancer detection and treatment.

[0069] 18 shows another conventional approach to prostate cancer detection and treatment, defined as an isotropic region of interest (ROI) expansion approach. This approach defines a uniform or isotropic boundary around the ROI. However, this approach does not take into account the unpredictable and asymmetrical manner in which tumor growth progresses, which is often not visible on MRI, demonstrating the need for a more comprehensive and patient-specific approach to cancer detection and treatment.

[0070] 19-28 show various visual representations and software and user interface implementations for mapping cancer using machine learning algorithms, as discussed above.

[0071] The articles “a,” “an,” and “the” are intended to mean the presence of one or more of the elements in a subsequent description. The terms “comprising,” “including,” and “having” are intended to be inclusive and mean that there may be additional elements other than the listed elements. Furthermore, it should be understood that references to “one embodiment” or “embodiments” of the present disclosure are not intended to exclude the existence of additional embodiments that also incorporate the recited features. Any numbers, percentages, ratios, or other values ​​set forth herein are intended to include that value and other values ​​that are “about” or “approximately” the recited value, as would be understood by one of ordinary skill in the art covered by the embodiments of the present disclosure. Accordingly, recited values ​​should be interpreted broadly enough to encompass values ​​that are at least sufficiently close to the recited value to perform the desired function or achieve the desired result. Recited values ​​at least include expected variations in suitable manufacturing or production processes, and may include values ​​that are within 5%, 1%, 0.1%, or 0.01% of the recited value.

[0072] Those skilled in the art will appreciate, in light of this disclosure, that equivalent structures do not depart from the spirit and scope of the disclosure, and that various changes, substitutions, and alterations can be made to the embodiments disclosed herein without departing from the spirit and scope of the disclosure. Equivalent structures, including functional "means-plus-function" clauses, are intended to cover structures described herein as performing the recited function, including structural equivalents that operate in the same manner and equivalent structures that provide the same function. It is the express intention of the applicant not to invoke means-plus-function or other functional claims in any claim, except where the words "means for" appear with the associated function. Each addition, deletion, and modification to the embodiments that comes within the meaning and scope of the claims is intended to be protected by the claims.

[0073] As used herein, the terms "approximately," "about," and "substantially" refer to an amount close to the stated amount that still performs a desired function or achieves a desired result. For example, the terms "approximately," "about," and "substantially" can refer to an amount that is within less than 5%, less than 1%, less than 0.1%, and less than 0.01% of the stated amount. Furthermore, any directions or reference frames in the foregoing description should be understood to be relative directions or movements only. For example, any references to "top" and "bottom" or "upper" or "lower" merely describe the relative positions or movements of the associated elements.

[0074] The present invention may be embodied in other specific forms without departing from its spirit or essential characteristics. The described embodiments are to be considered in all respects only as illustrative and not restrictive. The scope of the invention is, therefore, indicated by the appended claims rather than by the foregoing description. All changes that come within the meaning and range of equivalency of the claims are intended to be protected within their scope.

Claims

1. 1. A device for mapping cancer, comprising: a processor electrically coupled to a memory component that stores electronic instructions that, when executed by the processor, cause the device to execute a machine learning algorithm configured to receive an input and generate an output based on the input; the input includes data elements from a medical image; The output includes an estimate of the probability of clinically significant cancer in each voxel of the three-dimensional image. device.

2. the input further comprises prostate-specific antigen (PSA); The probability of clinically significant cancer comprises the probability of clinically significant prostate cancer (csPCa). The device of claim 1 .

3. The device of claim 1 , wherein the output further comprises a cancer estimation map (CEM).

4. The device of claim 3 , wherein the CEM displays a color-coded heat map representing the probability of cancer in each voxel of the three-dimensional image.

5. The device of claim 4 , wherein the medical image is an MRI image of a patient's anatomy.

6. The device of claim 5 , wherein the anatomical structure comprises a prostate.

7. the CEM includes a lesion outline representing a lesion size of a cancerous lesion shown in the three-dimensional image; The lesion contour includes an encapsulation confidence score The device of claim 4.

8. The device of claim 7 , wherein the output further comprises a visual curve representing the encapsulation confidence score versus the lesion size.

9. The device of claim 8 , wherein the visual curve includes points representing particular lesion sizes and particular encapsulation confidence scores.

10. 10. The device of claim 9, wherein the points are configured to be visually manipulated along the visual curve to vary the particular lesion size and the particular encapsulation confidence score represented by the points.

11. The device of claim 10 , wherein manipulation of the points alters the lesion contour.

12. The device of claim 1 , wherein the medical image comprises an MRI image.

13. 1. A method for mapping cancer, comprising: inputting data elements from the medical image into a machine learning model that estimates the probability of clinically significant cancer in the patient; and generating an output via the machine learning model that includes an estimate of the probability of clinically significant cancer in each voxel of the three-dimensional image. A method comprising:

14. 14. The method of claim 13, further comprising inputting data elements from a biopsy and a biopsy pathology label into the machine learning model.

15. The method of claim 13 , wherein the machine learning model is trained based on a population dataset that includes the data elements.

16. 16. The method of any one of claims 13 to 15, wherein the output comprises a visual representation of the three-dimensional image having a color-coded heat map representing the probability of clinically significant cancer at each voxel.

17. 1. A method for mapping cancer, comprising: inputting data elements from the medical image into a machine learning model that estimates the probability of clinically significant cancer in the patient; and displaying a visual representation of the probability at each voxel of the three-dimensional image, the visual representation comprising: a cancer estimation map (CEM) showing a color-coded heat map representing the probability of clinically significant cancer overlaid on the image, the CEM including a lesion outline representing the size of the cancerous lesion; and a curve representing the encapsulation confidence score versus the size, the curve including points representing the size of the lesion and the encapsulation confidence score; the points are configured to be visually manipulated along the curve to vary the size of the lesion represented by the points and the encapsulation confidence score; Manipulation of the points changes the lesion contour. A method comprising:

18. 18. The method of claim 17, further comprising displaying an interventional device in position relative to the image.

19. 20. The method of claim 18, wherein the position of the interventional device is configured to be changed relative to the image.

20. 20. The method of claim 18 or 19, further comprising displaying the location of the biopsy core overlaid on the image.

Images