A system and related methods to support the diagnosis of advanced prostate cancer

Abstract

The present invention relates to a system and method for computer-aided detection (CAD) of aggressive prostate cancer. Automated diagnosis of aggressive prostate cancer is difficult, and methods that work on some MRI scanners do not work well when performed on images from a different scanner. By using machine learning techniques to study a database containing MRI data and biopsy results of 265 patients taken with four different types of MRI scanners, the inventors established a method to automatically determine the presence of aggressive cancer, and when the method was performed on another database containing MRI data of 270 patients taken with a different MRI scanner than the first database, it showed high sensitivity and specificity. The method involves providing the MRI images to a system that calculates a score for a portion of the prostate gland and determining that portion contains aggressive cancer based on whether a criterion dependent on the score is verified.

Description

[Technical Field]

[0001] The present invention relates to a system for assisting in the diagnosis of advanced prostate cancer, and to a method for assisting in the diagnosis of advanced prostate cancer.

[0002] With over one million new cases reported annually, prostate cancer is the most common cancer in men worldwide and the second leading cause of cancer-related death. Diagnosis relies on ultrasound-guided biopsy. However, ultrasound often fails to detect tumors, and its role is limited to guiding the biopsy needle into the prostate.

[0003] Typically, physicians perform 12 systematic biopsies. However, clinically significant tumors are often missed during these biopsies. As a result, patients may not receive appropriate treatment.

[0004] In other patients, when the actual tumor is small and mild and does not threaten the patient's health, the biopsy needle may accidentally enter an isolated site of cancer cells. As a result, these patients undergo invasive treatment, which may result in serious side effects such as incontinence or impotence, without providing any medical benefit.

[0005] To improve the targeting accuracy of biopsies, several image processing techniques have been proposed that attempt to recognize signs of advanced cancer in MRI images. Examples include the paper by Au Hoang Dinh et al., "Quantitative Analysis of Multiparametric MR Images of the Prostate for the Detection of Advanced Prostate Cancer in the Peripheral Region: A Multi-Imager Study," published in 2016 in Volume 280, Number 1, page 117 of the journal *Radiology*, and "Characterization of Prostate Cancer with a Gleason Score of 7 or Higher Using Quantitative Multiparametric MR Images: Validation of a Computer-Assisted Diagnostic System in Patients Applied to Prostate Biopsies," published in 2018 in Volume 287, Number 2, page 525 of the journal *Radiology*.

[0006] However, known methods remain less reliable for detecting advanced cancer than those performed by experienced radiologists. In addition, these existing methods are typically developed based on MR images acquired from a limited number of MR scanners, and their reliability is significantly reduced when they are used to detect cancer on MR images acquired using a different type of MR scanner than the one used in the original base.

[0007] Therefore, there is a need for a system that can help radiologists diagnose advanced prostate cancer, is more robust than existing systems, reliably detect advanced prostate cancer, and, in particular, reliably work with MR images acquired using different types of MR scanners. [Overview of the Initiative]

[0008] overview From this perspective, this specification relates to a system for assisting in the diagnosis of advanced prostate cancer as described in claim 1.

[0009] In a specific embodiment, the system is as described in any one of claims 2 to 11.

[0010] This specification also relates to a method for assisting in the diagnosis of advanced prostate cancer as described in claim 12.

[0011] This specification also refers to the computer described in claim 13. In the program as well To relate to.

[0012] This specification also relates to the information medium described in claim 14.

[0013] This specification further relates to a method for diagnosing advanced prostate cancer, comprising steps for carrying out the method according to claim 12. [Brief explanation of the drawing]

[0014] The features and advantages of the present invention will become apparent from the following specification and are shown only as non-limiting examples, and are prepared with reference to the relevant drawings. [Figure 1] This is a schematic diagram of a system that assists in the diagnosis of advanced prostate cancer. [Figure 2] Figure 1 shows a flowchart of each step as an example of how the system can be used to assist in the diagnosis of advanced prostate cancer. [Figure 3] This graph shows the unique characteristics of the system, as an example, as a function of several parameters, as shown in Figure 1. [Figure 4] This graph shows the unique characteristics of the system, as another example, as a function of several parameters, as shown in Figure 1. [Figure 5] This graph shows the unique characteristics of the system as yet another example, as a function of several parameters, as shown in Figure 1. [Modes for carrying out the invention]

[0015] Figure 1 shows a schematic diagram of system 10 that assists in the diagnosis of advanced prostate cancer.

[0016] System 10 includes a control module 15, a human-machine interface 20, and a computer. Program 22 Includes.

[0017] More generally, system 10 is an electronic computer capable of processing and / or converting, transmitting, or storing data, which is illustrated as electronic or physical quantities in the registers and / or memory of system 10, into other similar data corresponding to physical data in memory, registers, or other types of displays.

[0018] System 10 is configured to perform a method to assist in the diagnosis of advanced prostate cancer.

[0019] The control module 15 is configured to receive at least one magnetic resonance image of the subject's prostate from a separate device 25, such as a magnetic resonance imaging scanner 25.

[0020] The control module 15 includes, for example, a data processing unit such as a processor 35, a memory 40, and a reader 45 for the readable information medium 50.

[0021] computer· Program 22 This includes information media 50.

[0022] The readable information medium 50 is a medium that is readable by the data processing unit 35. The readable information medium 50 is a medium suitable for storing electronic instructions in memory and can be coupled to the bus of the computer system.

[0023] As an example, the readable information medium 50 is an optical disc, CD-ROM, magneto-optical disc, ROM memory, RAM memory, EPROM memory, EEPROM memory, magnetic card, or optical card.

[0024] The readable information medium 50 stores a program, including program instructions, in its memory.

[0025] A computer program can be loaded onto a data processing unit 35, and when the computer program is applied onto a processor 35, it is adapted to apply a method for assisting in the diagnosis of advanced prostate cancer.

[0026] The human-machine interface 20 is configured to enable the transmission of information between a human, such as a doctor, operating the system 10 and the control module 15.

[0027] The human-machine interface 20 includes a display screen, such as a touchscreen. Optionally, the human-machine interface 20 further includes a mouse and / or a keyboard and / or speakers and / or a microphone.

[0028] The scanner 25 is, for example, separate from the system 10, and is located in a different room or building, and is connected to the system 10 by a wired or wireless network.

[0029] In a possible modification, the scanner 25 is part of the system 10. For example, a control module can instruct the scanner 25 to acquire a magnetic resonance image(s).

[0030] In another variation, the separate device 25 is a server or database where magnetic resonance images (or more) are stored, and in this case, the magnetic resonance images are acquired, for example, by one or more scanners and transmitted to the separate device 25 via a network.

[0031] This scanner is a multiparametric magnetic resonance imaging ("MRI") scanner that can acquire images of a subject's organs using different MRI techniques, particularly by measuring T1-weighted and T2-weighted MRI signals emitted from the organs, as shown below.

[0032] Next, the computer. Program The operation of system 10 in the interaction will be explained with reference to Figure 2, which illustrates an exemplary application of a method to assist in the diagnosis of advanced prostate cancer.

[0033] A method to assist in the diagnosis of advanced prostate cancer includes an acquisition step 100, a transmission step 110, a selection step 120, a calculation step 130, a determination step 140, and a signaling step 150.

[0034] During acquisition step 100, at least one magnetic resonance image, for example, a set of magnetic resonance images, is acquired by scanner 25.

[0035] Each image is a picture of the subject's prostate.

[0036] The subjects are, for example, patients exhibiting symptoms likely to be caused by prostate cancer. In the modified case, the subjects exhibit no symptoms whatsoever.

[0037] Each image represents at least a portion of the peripheral region ("PZ") of the subject's prostate. In particular, each image shows at least a portion of the prostatic zone (PZ) and at least a portion of the transitional zone (TZ) of the prostate.

[0038] A set of magnetic resonance images (or images) includes at least one first image and optionally one or more second images.

[0039] Each image contains an image element or a set of "pixels".

[0040] The first image described above, or each of the first images, is, for example, an apparent diffusion coefficient mapping. Such a mapping is a two-dimensional representation of the apparent diffusion coefficient values ​​of the imaged regions of the prostate. In such a mapping, each pixel contains the apparent diffusion coefficient value of the corresponding region of the prostate.

[0041] The apparent diffusion coefficient is a measure of the degree of diffusion of water molecules within a tissue and can be obtained by diffusion-weighted imaging (DWI). The apparent diffusion coefficient value is calculated from a set of a first T2*-weighted image of the prostate and at least two diffusion-weighted images, a method that is known in itself. Each diffusion-weighted image corresponds to a spatial direction and is a mapping of the T2* signal attenuated as a function of how easily water diffuses in the corresponding direction.

[0042] The apparent diffusion coefficient is given by the amount of air per square millimeter per second (mm²).2 It is expressed as / s).

[0043] At least one second image is, for example, a T1-weighted image containing a set of pixels, where each pixel is associated with a value of the intensity of the T1-weighted signal originating from a corresponding region of the prostate. In particular, a set of consecutive T1-weighted images is acquired over a predetermined time period.

[0044] The aforementioned predetermined time is the time it takes for the contrast agent to enter the prostate gland or to exit the prostate gland. In other words, the volume concentration of the contrast agent increases or decreases monotonically during that predetermined time.

[0045] For example, when a contrast agent is injected into a subject, the concentration of the contrast agent increases in the prostate, particularly in each first or second region, until it reaches a maximum value, and then decreases as the contrast agent is discharged from the prostate. The T1-weighted signal intensity in each region of the prostate is a function of the contrast agent concentration and increases significantly as a function of the contrast agent concentration, but this increase is nonlinear. Therefore, the T1-weighted signal intensity, like the contrast agent concentration, tends to increase over time and then decrease.

[0046] Images are acquired during a predetermined time period after injection, during which the contrast agent enters the prostate. In a modified version, images are acquired during a predetermined time period following the moment when the concentration of the contrast agent in the prostate reaches its maximum, during which the contrast agent is discharged from the prostate after reaching its maximum concentration.

[0047] The contrast agent is, for example, gadolinium.

[0048] During the transmission step 110, all first or second images are transmitted to the control module 15.

[0049] For example, all first or second images are acquired by the scanner 25 and immediately transmitted to the control module 15 via the network.

[0050] In a possible modification, the first or second image is transmitted by the scanner 25 to a database or server, stored on the database or server for a period of time, and after the storage time has elapsed, it is transmitted to the control module 15, for example, when a physician decides to check whether the subject's prostate contains advanced cancer. In this case, the method for assisting in the diagnosis of advanced prostate cancer can be considered to not include the acquisition step 100.

[0051] During selection step 120, at least one first part of PZ is selected.

[0052] For example, at least one image, e.g., a first image (or more), is presented to the physician on the human-machine interface 20, and the physician outlines a portion of the first image that the physician considers to be of interest, in particular a portion of the PZ that the physician considers to potentially contain advanced cancer from an image perspective.

[0053] Optionally, a set of different first parts of the PZ is selected, for example, by a physician, by outlining multiple regions of the first image.

[0054] In possible modifications, the control module 15 automatically divides an image, such as the first image or one of the first images, into multiple parts by, for example, superimposing a square grid onto the image, and the control module 15 considers each region of PZ separated from each other by the grid as one of those first parts.

[0055] Optionally, at least one secondary portion of the prostate is selected. Each secondary portion is a part of the TZ.

[0056] Each second part is either outlined by a doctor, for example, or automatically selected in the same way as the first part(s).

[0057] During calculation step 130, the control module 15 calculates a first score P based on at least one first image(s) received. Optionally, as detailed below, the control module further calculates a second score Y from at least one first image(s) received.

[0058] In particular, the first score P is calculated for each selected first part, and / or the second score Y is calculated for each selected second part. It will be calculated.

[0059] Each first score P is a function of at least one of the first quantity x1, the second quantity x2, and the third quantity x3.

[0060] The first quantity x1 is calculated for the first part under consideration according to the following formula:

number

[0061] The normalized wash-in rate for each region is calculated by generating a curve of the intensity of the T1-weighted signal for a region as a function of time from the intensity values ​​of the pixels corresponding to the region in a second consecutive image, using a known method, and then estimating the slope of that curve.

[0062] The normalized wash-in or wash-out rate is estimated, in particular, from the normalized T1-weighted signal, where the intensity of the T1-weighted signal of a pixel at a given time is divided by the intensity of the T1-weighted signal of the same pixel before the contrast agent enters the prostate, for example, before the injection of the contrast agent or before the contrast agent reaches the prostate (i.e., before the start of the intensity increase).

[0063] It should be noted that different normalization methods can be used, and they are equivalent, because the corresponding values ​​of parameter a1 and wash-in rate or cash-hour rate for one normalization method can be estimated from the values ​​for another normalization method.

[0064] In possible modifications, instead of dividing all T1-weighted signal values ​​by the value of this signal before contrast agent injection, the wash-in or wash-out velocity is calculated from the gradient expressed as a percentage of the original value.

[0065] In this modified example, a1 is mm 2 Instead of expressing it as mm, 2 It can be expressed as a percentage, and therefore the value of a1 shown in 4 is mm 2 This is 100 times smaller than when expressed using the other method.

[0066] When implementing the present invention, it should be considered that all types of normalization of the wash-in or wash-out speed can be used indiscriminately, along with adaptations corresponding to the value of a1 and the units used to represent the wash-in or wash-out speed.

[0067] The gradient of the increasing portion of the curve, particularly the maximum gradient (corresponding to the gradual increase in contrast agent concentration in the region under consideration), is the normalized wash-in velocity of that region. In possible variations, the normalized wash-in velocity of the region is the average gradient of the increasing portion of the curve.

[0068] A wash-in or wash-out rate is referred to as a "normalized" wash-in or "normalized" wash-out if the intensity curve is constructed using normalized intensity values, i.e., values ​​obtained by dividing the T1-weighted signal of a single pixel or the intensity of a region in a second image at a different time by the intensity of the T1-weighted signal of the same pixel or region before the contrast agent is injected. Such normalization allows for easy comparison of wash-in or wash-out rates obtained from different MRI devices.

[0069] It should be noted that, as will be discussed later, other normalization methods can also be used.

[0070] The gradient of the decreasing portion of the curve (corresponding to the gradual decrease after the contrast agent concentration reaches its maximum in the region under consideration), such as the average gradient or maximum gradient, is the normalized washout rate for that region. In practice, the normalized washout rate is often the average gradient.

[0071] If parameter W is a normalized wash-in rate, then parameter W is, for example, the arithmetic mean of the normalized wash-in rates of the region corresponding to each pixel of the first part.

[0072] If parameter W is a normalized washout rate, then parameter W is, for example, the arithmetic mean of the normalized washout rates of the region corresponding to each pixel of the first part.

[0073] In a modified example, the normalized wash-in or normalized wash-out rate is calculated by, for each first portion of each second image, calculating the average value of the T1-weighted signals of all pixels within the first portion and generating a normalized curve of the average value of the first portion calculated as a function of time. The generated time curve has the same overall shape as that of each pixel because the average value of the contrast agent concentration in the first portion considered increases and then decreases with time. In this case, the wash-in rate value is the slope of the increasing portion of the generated time curve, in particular the maximum or average slope, and the normalized wash-out rate is the slope of the decreasing portion of the curve, in particular the maximum or average slope.

[0074] It should be noted that the normalized wash-in or wash-out rate can also be calculated for the second part(s) in the same manner as described above.

[0075] The parameter ADC is calculated by the control module 15 by arranging the apparent diffusion coefficient values ​​of the pixels corresponding to the first part under consideration in increasing order, and the parameter ADC is the percentile of the apparent diffusion coefficient value of the first part.

[0076] "Percentile" means that the parameter ADC is the value of the apparent diffusion coefficient that appears at a given position within a set of arranged apparent diffusion coefficient values.

[0077] For example, the tenth percentile of the apparent diffusion coefficient is the apparent diffusion coefficient such that 90 percent of the apparent diffusion coefficients are exactly above the tenth percentile, and 10 percent of the diffusion coefficients are below the tenth percentile.

[0078] The second percentile is one in which 98 percent of the apparent diffusion coefficient values ​​are strictly better than the second percentile, and 2 percent of the diffusion coefficient values ​​are inferior to or equal to the second percentile.

[0079] If the number of pixels corresponding to the portion under consideration is not divisible by 100, it should be noted that the percentile can be calculated by interpolation between the two pixels that contain that percentile.

[0080] Unless otherwise specified, the values "included between" two limit values correspond to the range that includes both limit values.

[0081] ADC values are, in many cases, expressed in square millimeters per second (mm 2 / s).

[0082] When W is the normalized wash-in rate, the first constant a1 is, for example, between -6×10 -3 mm 2 and -10 -3 mm 2 In some embodiments, the first constant a1 is between -5×10 -3 mm 2 and -1.5×10 -3 mm 2 For example, between -5×10 -3 mm 2 and -3×10 -3 mm 2 is included.

[0083] In this specification, the symbol "×" between two numbers indicates multiplication.

[0084] The parameter ADC is, particularly when W is the normalized wash-in rate value, the percentile below 40, particularly the percentile below 30, particularly the percentile below 10.

[0085] When W is the normalized wash-out rate, the first constant a1 is, for example, between 2× -2 mm 2 and 15×10 -2 mm 2 is included.

[0086] The parameter ADC indicates, for example, that the percentile is 15 or less when W is a normalized washout rate value.

[0087] The second quantity x2 is calculated for the first part considered according to the following formula:

number

[0088] The second constant b1 is, for example, 10 -7 mm 2 / square seconds and 14 × 10 -7 mm 2 / square second (mm 2 / s 2 ) is included in between. In some cases, the second constant b1 is 10 -7 mm 2 / s 2 and 9x10 -7 mm 2 / s 2 It is included in between.

[0089] The time-to-peak value is the time between the first and second moments on the curve of the T1-weighted signal value as a function of time. The first moment is the time after the injection of contrast agent into the subject when the T1-weighted signal value begins to increase (i.e., the moment when the injected contrast agent begins to enter the first portion of the curve). The second moment is the time of the maximum concentration of contrast agent in the first portion and corresponds to the maximum intensity of the T1-weighted signal.

[0090] The time value to the peak can be obtained, for example, by calculating the average value of the T1-weighted signals of all pixels for each second image, generating a curve calculated as a function of time based on that average value, and measuring the time value to the peak on that curve.

[0091] In the modified example, the time to peak value is the average of the time to peak values ​​calculated for each pixel corresponding to the first part.

[0092] The third quantity x3 is calculated for the corresponding first part according to the following formula:

number

[0093] The third constant c0 falls between, for example, 2.41 and 6.44, and especially between 3.5 and 5.0. In particular, the third constant c0 is equal to 4.2.

[0094] The fourth constant c1 is -7570 seconds / mm 2 and -4020 seconds / mm 2 During this period, especially at -7150 seconds / mm 2 and -5270 seconds / mm 2 It is included in the period.

[0095] The first score P for each first part is calculated from at least one of the first quantity x1, the second quantity x2, and the third quantity x3.

[0096] The first score P is equal to one of the following quantities, for example: the first quantity x1, the second quantity x2, and the third quantity x3.

[0097] The first score P is calculated using logistic regression in possible variations.

[0098] In statistics, the logistic regression model is used to model the probability of an event occurring, for example, to model the probability that the first part of a sequence includes advanced cancer.

[0099] The first score P is calculated, for example, using logistic regression according to the following formula.

number

number

[0100] In this case, score P is the probability that the first part considered includes advanced cancer.

[0101] For example, in our data, the score is,

number

[0102] In another modified example, the first score P is obtained by calculating two or three probabilities, each of which is calculated using logistic regression according to Equation 4 for each of the first quantity x1, the second quantity x2, and the third quantity x3, and then obtained by summing the three calculated probabilities.

[0103] In general, it should be noted that many types of scores P are conceivable if there is a bijective relationship between the calculated score P and any of the first quantity x1, the second quantity x2, and the third quantity x3, or any combination of these quantities (such as the sum or weighted sum).

[0104] For example, the score P is obtained by dividing or multiplying one of the first quantity x1, the second quantity x2, and the third quantity x3 by a constant, and in some cases, it may be a dimensionless quantity x, which is arbitrarily entered into equation 5 to calculate the score P.

[0105] As mentioned above, the units used to represent the parameters ADC, WI, WO, and TTP involved in the calculation of the first quantity x1, the second quantity x2, and the third quantity x3 can also be changed, and parameters a and b1 can be adapted accordingly.

[0106] Optionally, during calculation step 130, the control module 15 further calculates a second part of TZ, or a second score Y for each second part.

[0107] The second score Y is calculated as a function of the fourth quantity y, which is dimensionless.

[0108] The fourth quantity, y, is calculated as a function of the following equation.

number

[0109] The fifth constant d0 lies between 2.25 and 13.48, and is in particular equal to 7.9.

[0110] The sixth constant d1 is -23820 s / mm 2 and -4490s / mm 2 It falls within the range, particularly between -15030 and -6910. In particular, the sixth constant d1 is equal to -10740.

[0111] The second score, Y, is calculated using, for example, the following formula:

number

[0112] However, as with the case of the first score P, many different methods can be used to calculate the second score Y from the fourth quantity y. For example, the second score Y is equal to the fourth quantity y.

[0113] During the decision step 140, the control module 15 determines, for each first part, whether the first part of the PZ under consideration contains advanced cancer.

[0114] In particular, the control module 15 determines whether the first criterion based on the first score P has been verified, and if the criterion has been verified, it determines that the first part includes advanced cancer.

[0115] According to one embodiment, the criterion is that the first score P is less than or equal to the first threshold, in which case the control module compares the first score P with the first threshold and determines that the first portion includes advanced cancer if the first score P is less than or equal to the first threshold. If the first score P is strictly above the first threshold, the control module 15 determines that the first portion does not include advanced cancer.

[0116] This is particularly evident when the first score P is equal to one of the first quantity x1, the second quantity x2, and the third quantity x3. In this case, the first threshold is referred to, for example, as the "score threshold."

[0117] The score threshold is, for example, -0.8 × 10 if the first score P is equal to the first quantity x1. -3 mm 2 / s and 2.3 × 10 -3 mm 2 It is constructed between / s, where W is the normalized wash-in rate. In particular, the score threshold is 0.55 × 10 -3 mm 2 / s and 0.8 × 10 -3 mm 2 It is included within the / s section.

[0118] If W is the normalized washout rate, the score threshold is, for example, -1.1 × 10⁻¹⁰ -3 and 3.2 × 10 -3 mm 2 It is included within / s. In particular, the score threshold is 0.85 × 10 -3 mm 2 / s and 1.1 × 10 -3 mm 2 It is included within the / s section.

[0119] If the first score P is equal to the second quantity x2, then the score threshold is, for example, -1.1 × 10⁻¹⁰. -3 mm 2 / s and 3.3 × 10 -3 mm 2 During / s, especially 0.9 × 10 -3 mm 2 / s and 1.1 × 10 -3 mm 2 It is included within the / s section.

[0120] The score threshold is equal to 0.40, for example, if the score P is equal to the third quantity x3.

[0121] It should be noted that other criteria may also be considered.

[0122] When the first score P is calculated using equations 4 and 5, the criterion is that the first score P is greater than or equal to the first threshold. In this case, the control module compares the first score P to the first threshold and determines that the first portion includes progressive cancer if the first score P is greater than or equal to the first threshold. If the first score P is strictly below the first threshold, the control module 15 determines that the first portion does not include progressive cancer. The first threshold is referred to, for example, as the "probability threshold".

[0123] During the decision step 140, the control module 15 further determines, for each second portion, whether the second portion of the TZ under consideration contains advanced cancer.

[0124] In particular, the control module 15 checks whether the second criterion is validated, and if the criterion is validated, it determines that for each second portion, the second portion of the TZ considered includes advanced cancer. If the criterion is not validated, the control module 15 determines that the second portion does not include advanced cancer.

[0125] According to one embodiment, the control module compares a second score Y with a second threshold and determines that the second portion contains advanced cancer if the second score Y is less than or equal to the second threshold. In this case, the criterion is the fact that the second score Y is less than or equal to the second threshold. However, other criteria may be considered.

[0126] For example, the control module 15 calculates the probability that the second part contains advanced cancer, particularly using equation 6, and determines that the second part contains advanced cancer if that probability is greater than or equal to the second probability threshold.

[0127] The second probability threshold is, for example, between 0.70 and 0.77 when the value ADC used in Equation 6 is the 25th percentile of ADC.

[0128] If the control module 15 determines that one of the first and second parts (or more) contains advanced cancer, the control module 15 generates a message directed to the physician during the signaling step 150.

[0129] This message is intended to be sent to a physician via the human-machine interface 20.

[0130] This message informs the doctor that it has been determined that at least one of the first or second parts contains advanced cancer.

[0131] This message includes, for example, a representation of advanced cancer, or a first or second portion(s) containing advanced cancer, such as one of the first or second images, where the first or second portion containing advanced cancer is outlined in red, or accompanied by a written message stating that the first or second portion under consideration is likely to be cancerous.

[0132] This message can optionally be accompanied by an alarm-like sound indicating the detection of cancer.

[0133] It should be noted that this message can be in any format as long as the doctor can understand the information.

[0134] Subsequently, the physician may examine the first or second portion indicated in the message in detail on the first or second image, or by other means, to determine whether the physician shares the control module's assessment regarding the presence of advanced cancer in the first or second portion.

[0135] Furthermore, this method allows the physician who examined the first and second images to have the system 10 reverse-check the analysis of the subject's prostate, thereby enabling the diagnosis of cancers that may have been overlooked by the physician, if the system is instructed to draw the physician's attention to areas of the prostate where cancer may be present or where cancer may not be clearly visible.

[0136] In possible modifications, the signaling step 150 is not performed, and the fact that the control module 15 has determined the presence of advanced cancer is simply stored, for example, in memory 40.

[0137] A method to assist in the diagnosis of advanced prostate cancer includes, for example, a diagnostic step in which prostate cancer is diagnosed, which is performed as part of a prostate cancer diagnostic method and functions as a message(s) generated during the method to assist in the diagnosis of advanced prostate cancer.

[0138] In particular, the diagnostic methods for prostate cancer are part of the treatment methods for prostate cancer, and include the step of treating the prostate cancer diagnosed by performing the diagnostic methods for prostate cancer.

[0139] Formulas 1 to 3 and 5 were determined by the inventors during a study using a learning database, including MRI sequences, from 265 patients who were diagnosed with advanced prostate cancer and underwent treatment, including biopsy, multiparametric MRI, and radical prostatectomy.

[0140] T2-weighted, diffusion-weighted, and dynamic contrast-enhanced images were systematically recorded and prospectively reviewed by two independent radiologists. In each sequence, areas suspected of being cancerous were segmented. A computer program extracted 23 quantitative parameters from these areas.

[0141] The training database contains MR images acquired using different magnetic field values ​​from several MR scanners of different manufacturers and types. - Siemens Symphony scanner (1.5 Tesla, 63 patients, 233 suspected disease variables), - Discovery MR 750 scanner manufactured by General Electric Medical Systems (3T, 124 patients, 272 disease samples) - Achieve scanner manufactured by Philips Medical Systems (3T, 52 patients, 173 disease samples). - Philips Healthcare Ingenia scanner (3T, 26 patients, 56 disease samples)

[0142] The model was discovered using machine learning techniques, recursively browsing different variables and testing the performance of each variable one by one. A new variable was retained if its corresponding coefficient and the deviation it explains were statistically significant within the model and its correlation with each other variable in the model was less than 50%.

[0143] The threshold was determined using an intermediate database containing MRI images of prostates, including suspected lesions whose benign or malignant nature had been confirmed by biopsy, and was selected to achieve a 90% sensitivity.

[0144] The intermediate database, like the learning database, uses MR data obtained from GE's MR750 and Philips Healthcare's Ingenia scanners, with data from 101 and 11 patients respectively.

[0145] Equations 1-3 and 5 were constructed from the selected parameters, and their performance was then evaluated on a test database containing data (MRI images and biopsies) from patients different from the training and intermediate databases. The training database contained data only from patients with confirmed prostate cancer, while the test database included patients suspected of having prostate cancer but without confirmed cancer.

[0146] The method to assist in the diagnosis of prostate cancer is performed on MRI images in a test database and involves two radiologists outlining a first and second portion of the MRI image.

[0147] From data of 158 patients, radiologists identified a total of 238 suspicious lesions. Biopsies revealed 126 benign lesions (114 in the peritoneal zone and 12 in the transthermal zone), 34 cancers with a Gleason score of 6 (30 in the peritoneal zone and 4 in the transthermal zone), and 78 cancers with a Gleason score of 7 or higher (71 in the peritoneal zone and 7 in the transthermal zone).

[0148] The MR images in the test database were acquired using four different MR scanners. - Discovery MR750 scanner manufactured by General Electric Medical Systems (3T, 62 patients, 93 disease samples) - Philips Healthcare Ingenia scanner (3T, 22 patients, 41 lesions) - GE Medical Systems' Optima MR450w (1.5T, 66 patients, 91 disease cases), and - Ingenia (1.5T, 8 patients, 13 lesions) manufactured by Philips Healthcare.

[0149] The sensitivity and specificity of the systems using Equations 1, 2, 3, and 5 were estimated by comparing them with diagnostic results obtained from biopsies of the corresponding portions of the patient's prostate. The performance of the systems was measured from the specificity achieved when the sensitivity was set to 90% (referred to as "specificity sp90").

[0150] The accuracy of system 10 using Equation 1, when W is used as the normalized wash-in rate on the test database, consistently shows a singularity greater than 0.5, and the ADC is in the percentile of 40 or less, and the first constant a1 is -6 × 10⁻⁶ -3 mm 2 and -10-3 mm 2 It is included in between. Figure 3 shows the values ​​of the singularity sp90 as a function of ADC and a1 for system 10 using Equation 1 in the test database, where W is the normalized wash-in rate.

[0151] Figures 3-5 show contours that delineate regions with sp90 specificity values ​​higher than the threshold. For example, the line labeled "0.55" includes all regions of data corresponding to sp90 values ​​higher than 0.55.

[0152] In particular, if ADC is in a percentile less than or equal to 30, and the first constant a1 is -5 × 10 -3 mm 2 and -3 × 10 -3 mm 2 When the values ​​fell within the range, the specificity consistently exceeded 0.55. ADC is a percentile less than or equal to 10, and the first constant a1 is -4.5 × 10 -3 mm 2 and -3 × 10 -3 mm 2 When the values ​​fell within the range, the specificity was effectively greater than or equivalent to 0.58.

[0153] In the test database, ADC is the second percentile, and the first constant a1 is -3.96 × 10⁻⁵ mm 2 The best performance was achieved when the value was equal to 0.606 and the singularity was equal to 0.606.

[0154] The accuracy of system 10 using Equation 1 is such that, with W as the normalized washout rate, the ADC is in the percentile less than or equal to 15, and the second constant a1 is 2 × 10⁻⁶. -2 and 15×10 -2 mm 2 When used on the training database included in the period, it consistently showed a singularity greater than 0.45. Figure 4 shows the value of the singularity sp90 as a function of ADC and a1, where W is the normalized washout rate, in the training database of system 10 using Equation 1.

[0155] In particular, when the ADC is a percentile included between 2 and 10 and the first constant a1 is between 4×10 -2 mm 2 and 12×10 -2 mm 2 the specificity was consistently a value exceeding 0.50.

[0156] The highest performance was obtained on the learning database when the ADC was the second percentile, the first constant a1 was 10×10 -2 mm 2 and the specificity was equal to 0.549.

[0157] The accuracy of the system 10 using Equation 2, which was used on the test database, showed a specificity consistently exceeding 0.45 when the ADC was a percentile included between 20 and 45 and the second constant b1 was between 10 -7 mm 2 and 14×10 -7 mm 2 The value of the specificity sp90 of the system 10 using Equation 2, as a function of the ADC and b1, within the test learning database, is shown in FIG. 5.

[0158]

[0159] In particular, when the ADC was a percentile included between 20 and 45 and the second constant b1 was between 10 -7 and 9×10 -7 mm 2 the specificity was consistently a value exceeding 0.50.

[0160] The highest performance was obtained on the test database when the ADC was the 25th percentile, the second constant b1 was 5.1×10 -7 mm 2 and the specificity was equal to 0.539.

[0161] ​Examples of the first threshold and the determined constants c0 and c1 in the case where system 10 uses equation 4 and quantity x is different from the third quantity x3 are listed below.

[0162] a) When the ADC is in the smallest percentile, the first threshold falls between 0.18 and 0.37, the seventh constant c0 falls between 2.41 and 3.61, and the eighth constant c1 is -5880 s / mm 2 and -4020s / mm 2 The combination that gives the best results is one where the first threshold is equal to 0.28, the seventh constant c0 is equal to 2.9, and the eighth constant c1 is -4790 s / mm 2 (It was equal to [a certain value]).

[0163] b) When the ADC is in the second percentile, the first threshold is between 0.17 and 0.39, the seventh constant c0 is between 3.55 and 4.92, and the eighth constant c1 is -7150 s / mm 2 and -5270s / mm 2 The combination that gives the best results is one where the first threshold is equal to 0.27, the seventh constant c0 is equal to 4.17, and the eighth constant c1 is -6080 s / mm 2 (It was equal to [a certain value]).

[0164] c) When the ADC is in the ninth percentile, the first threshold falls between 0.17 and 0.38, the seventh constant c0 falls between 4.42 and 5.92, and the eighth constant c1 is -7570 s / mm 2 and -5660 s / mm 2 The combination that gives the best results is one in which the first threshold is equal to 0.27, the seventh constant c0 is equal to 5.09, and the eighth constant c1 is -6500 s / mm 2 (It was equal to [a certain value]).

[0165] d) When the ADC is in the tenth percentile, the first threshold is between 0.17 and 0.38, the seventh constant c0 is between 4.47 and 5.96, and the eighth constant c1 is -7560 s / mm 2 and -5650s / mm 2The combination that gives the best results is one where the first threshold is equal to 0.26, the seventh constant c0 is equal to 5.13, and the eighth constant c1 is -6490 s / mm 2 (It was equal to [a certain value]).

[0166] e) When the ADC is in the 25th percentile, the first threshold falls between 0.14 and 0.36, the seventh constant c0 falls between 4.59 and 6.44, and the eighth constant c1 is -7430 s / mm 2 and -5260s / mm 2 The combination that gives the best results is one in which the first threshold is equal to 0.27, the seventh constant c0 is equal to 5.49, and the eighth constant c1 is -6260 s / mm 2 (It was equal to [a certain value]).

[0167] Examples of the second threshold and the determined constants d0 and d1 in different cases where System 10 uses Equation 6 are listed below.

[0168] f) When ADC is in the second percentile, the second threshold is between 0.07 and 0.75, the ninth constant d0 is between 2.25 and 13.48, and the tenth constant d1 is -23820 s / mm 2 and -4490s / mm 2 The combination that gives the best result is one in which the second threshold is equal to 0.4, the ninth constant d0 is equal to 6.5, and the tenth constant d1 is -11520 s / mm 2 (It was equal to [a certain value]).

[0169] g) When the ADC is in the 9th percentile, the second threshold falls between 0.07 and 0.71, the 9th constant d0 falls between 3.83 and 12.5, and the 10th constant d1 is -19500 s / mm 2 and -6280s / mm 2 The combination that gives the best result is one where the second threshold is equal to 0.38, the ninth constant d0 is equal to 7.3, and the tenth constant d1 is -11320 s / mm 2 (It was equal to [a certain value]).

[0170] h) When the ADC is in the tenth percentile, the second threshold is between 0.06 and 0.71, the ninth constant d0 is between 3.96 and 12.51, and the tenth constant d1 is -19100 s / mm 2 and -6430s / mm 2 The combination that gives the best result is one where the second threshold is equal to 0.39, the ninth constant d0 is equal to 7.4, and the tenth constant d1 is -11360 s / mm 2 (It was equal to [a certain value]).

[0171] i) When the ADC is in the 25th percentile, the second threshold falls between 0.14 and 0.64, the ninth constant d0 falls between 4.9 and 11.19, and the tenth constant d1 is -15030 s / mm 2 and -6910s / mm 2 The combination that gives the best result is one in which the second threshold is equal to 0.4, the ninth constant d0 is equal to 7.9, and the tenth constant d1 is -10740 s / mm 2 (It was equal to [a certain value]).

[0172] System 10, using Equation 4 where quantity x is the first quantity x1 and ADC is the second percentile, was found to have the following AUC and specificity (the percentages of sensitivity and specificity when using contours drawn by other radiologists are shown in parentheses) when using contours drawn by one radiologist in the following cases. - Case a: AUC was between 76% and 86% (between 78% and 87%), specificity sp90 was between 33% and 52% (between 28% and 63%), - Case b: AUC was between 77% and 86% (between 79% and 87%), specificity sp90 was between 34% and 54% (between 29% and 66%), - Case c: AUC was between 78% and 88% (between 79% and 88%), specificity sp90 was between 37% and 61% (between 32% and 74%), - Case d: AUC was between 79% and 88% (between 79% and 88%), and specificity sp90 was between 37% and 62% (between 32% and 74%). - Case e: AUC was between 79% and 88% (between 80% and 89%), and specificity sp90 was between 39% and 62% (between 38% and 72%).

[0173] AUC is a performance quantified by the area under the curve of the true positive rate as a function of the false positive rate. Such curves are well-known in the field and are often referred to as the ROC (Receiver Operating Characteristic) curve.

[0174] System 10 using Equation 5 was found to have the following AUC and specificity (the percentages of sensitivity and specificity when using contours drawn by other radiologists are shown in parentheses) when using contours drawn by one radiologist in the following cases. - Case f: AUC was between 47% and 86% (between 35% and 83%), specificity sp90 was between 0% and 80% (between 0% and 82%), - Case g: AUC was between 46% and 88% (between 28% and 85%), specificity sp90 was between 0% and 83% (between 0% and 82%), - Case h: AUC is between 46% and 89% (between 30% and 85%), specificity sp90 is between 0% and 83% (between 0% and 82%). - Case i: AUC was between 50% and 94% (between 29% and 82%), and specificity sp90 was between 0% and 83% (between 0% and 75%).

[0175] It should be noted that both quantities x1 and x2 are a combination of the centile of the ADC and the dynamic quantity of the MRI image (i.e., a quantity that reflects the change in value as a function of time during the inflow or outflow of contrast agent into or out of the prostate), namely the wash-in rate, wash-out rate, or time-to-peak value.

[0176] Therefore, part of the present invention lies in the fact that the inventors have discovered a range of values ​​a1, b1, and ADC percentiles that reflect the possibility of cancer being present in a part of the prostate when such a combination is used, and that when used together, they lead to the efficient detection of cancer from images.

[0177] The computer-aided diagnostic methods described herein can be used to establish a clinical diagnosis based on the results of the diagnostic methods in order to aid in the diagnosis of advanced prostate cancer.

[0178] In some embodiments, the method of the present invention is performed in vitro or ex vivo.

[0179] To confirm the initial results of this method, the inventors conducted additional verification tests. These results are shown below as merely one example.

[0180] In this example, the evaluation was performed using a separate test dataset (named "external test dataset") independent of the test dataset mentioned above.

[0181] This new database contains images from 104 patients and 126 disease variables, all imaged using a single scanner, a GE Medical Systems Signa Voyager scanner (1.5T, 104 patients, 126 disease variables).

[0182] These 104 patients were suspected of having cancer and underwent mpMRI and subsequent biopsies at other facilities.

[0183] Similar to the previously mentioned study, PI-RADSv2 (Prostate Imaging-Reporting and Data System version 2) scores, which were prospectively assigned to each lesion at the time of biopsy, were retrieved from the patient's medical record. Next, based on the mpMRI report, two radiologists with 19 years of experience (OR, R1) and 5 years of experience (PCM, R2) blinded each other's biopsy findings and retrospectively outlined the target lesion at the slice level most likely to be representative in T2-weighted images, diffusion-weighted images, and DCE images. The lesion score was calculated using either a PZ or TZ model depending on its location. Thus, each lesion received two CAD scores: a PI-RADSv2 score (prospectively assigned at the time of biopsy) and a region of interest (ROI) outlined by R1 and R2. Subsequently, the PI-RADSv2 score and the score obtained according to this invention were compared with the biopsy findings.

[0184] The following table details the distribution of PI-RADSv2 scores and scores derived from the present invention as functions of biopsy results.

[0185] [Table 1]

[0186] In the table above, the data represents the number of patients. ISUP is an abbreviation for the International Society of Urological Pathology's grading system for measuring the severity of cancer. CAD is an abbreviation for the computer-aided diagnostic system according to the present invention.

[0187] In the new test dataset, the AUC scores using ROIs contoured with R1 and R2 were 82% (95% CI: 76-89) and 86% (95% CI: 79-93), respectively. These were not significantly different from the PI-RADSv2 scores assigned at the time of biopsy (85%, 95% CI: 79-91, p=0.82 and p=1, respectively).

[0188] This new test presented challenges because all patients were imaged with newer scanners, different from those used in the training dataset. Nevertheless, the system according to the present invention delivered robust results not only in terms of overall diagnostic performance (quantified by AUC) but also in terms of diagnostic thresholds (quantified by sensitivity and specificity). Surprisingly, the threshold that showed 90% sensitivity in the pre-test dataset showed similar sensitivity in the internal test dataset (85%–89%) and the external test dataset (92%). At this level of sensitivity, the specificity of the computer-aided diagnostic system according to the present invention was good (64%–76%), suggesting that it can offer a better sensitivity / specificity trade-off than the classical threshold of PI-RADSv2.

Claims

1. A system (10) for assisting in the diagnosis of progressive prostate cancer, wherein the system includes a control module (15), and the control module (15) is - Receiving at least one magnetic resonance image of at least a first portion of the peripheral region of the subject's prostate, wherein the image includes a set of pixels, and for each pixel, the image includes an apparent diffusion coefficient value for the corresponding region of the prostate. - Calculate the first score, and, - If the first criterion based on the first calculated score is verified, it is determined that the first portion includes advanced cancer; or if the first criterion is not verified, it is determined that the first portion does not include advanced cancer. It is configured in such a way, and The first score mentioned above is calculated by the following quantity x １ and quantity x ２ It is calculated as a function of at least one of the following: [Number 9] Here, W is the value of the normalized wash-in rate or the normalized wash-out rate of the first part, and the wash-in rate value or the wash-out rate value is expressed in s−1 and is estimated from the set of magnetic resonance images of the first part. ADC is the percentile of the apparent diffusion coefficient value of the first part, expressed in units of mm 2 2 / s, and a １ is a first constant. When W is the normalized wash-in rate value, the first constant is between −6×10 －３ mm ２ and −10 －３ mm ２ and the percentile is 40 or less. When W is the normalized wash-out rate value, the first constant is between 2×10 －２ mm ２ and 15×10 －２ mm ２ and the percentile is 15 or less. [Number 10] Here, b １ is 10 －７ mm ２ / s ２ and 14 x 10 －７ mm ２ / s ２ A second constant that falls between the first and second parts, where ADC is a percentile that falls between 20 and 45 of the apparent diffusion coefficient values ​​of all pixels corresponding to the first part, TTP is the time value to the peak of the first part, expressed in seconds, where the time value to the peak is the time duration from the first moment to the second moment, where the first moment is the arrival time of the contrast agent in the prostate, and the second moment is the time of the maximum contrast density of the MRI signal in the first part, in the system.

2. W is the normalized wash-in velocity value, and the first constant a １ is -5 × 10 －３ mm ２ and -3 × 10 －３ mm ２ The system according to claim 1, which is included between the two.

3. The system according to claim 2, wherein the percentile is between 0 and 30.

4. The second constant b １ is 10 －７ mm ２ / square seconds and 9 x 10 －７ mm ２ The system according to claim 1, which is included in the interval of / square seconds.

5. The system according to claim 1, wherein the ADC is the 25th percentile of the apparent diffusion coefficient value for all pixels corresponding to the first portion, when the first score is calculated as a function of the time value to the peak.

6. The system according to claim 1, wherein a wash-in velocity value or a wash-out velocity value is estimated for each pixel corresponding to the first portion, and W is the average value of the wash-in velocity value or wash-out velocity value estimated for each pixel.

7. The system according to claim 1, wherein a plurality of images of the first portion are acquired during a predetermined time, the predetermined time being a predetermined time during which a contrast agent enters the prostate or a predetermined time during which the contrast agent exits the prostate, each image includes, for each pixel, an intensity value of the T1-weighted signal for the corresponding region of the prostate, the control module is configured to calculate, for each image, the average of the normalized intensity values ​​of the pixel corresponding to the first portion, each normalized value being equal to the value obtained by dividing the measured value by the value of the same pixel before the contrast agent entered the prostate, the normalized wash-in velocity value is the upward slope of the curve of the average value calculated as a function of time, and the normalized wash-out velocity value is the downward slope of the curve.

8. The system according to claim 1, wherein the control module (15) is configured to compare the first score with a threshold and determine that the first portion includes progressive cancer if the first score is less than or equal to the threshold.

9. The first score is the first quantity x １ If W is equal to the normalized wash-in rate value, then the threshold is -0.8 × 10 －３ square millimeters per second and 2.3 × 10⁻⁶ －３ If the data is contained within a square millimeter per second, or if W is a normalized washout rate value, the threshold is -1.1 × 10⁻¹⁰ －３ Square millimeters per second and 3.2 × 10⁻⁶ －３ The system according to claim 8, which is included in a period of square millimeters per second.

10. The first score mentioned above is the second quantity x ２ Equal to, the threshold is -1.1 × 10 －３ square millimeters per second and 3.3 × 10⁻⁶ －３ The system according to claim 8, which is included in a period of square millimeters per second.

11. Each image includes an image of the second portion of the transitional region of the prostate, and the control module (15) further, - The second score Y is calculated as a function of the fourth quantity y, and the fourth quantity y follows the following equation: [Math 11] Here, d ０ is the fifth constant, and d １ The rate is -23820 seconds / mm ２ and -4490 seconds / mm ２ The sixth constant is included between , the fifth constant is included between 2.25 and 13.48, ADC is a percentile that is 25 or less of the apparent diffusion coefficient value of all pixels corresponding to the second portion, and - If the second criterion based on the calculated second score Y is verified, it is determined that the second portion includes advanced cancer; or, if the second criterion is not verified, it is determined that the second portion does not include advanced cancer. The system according to claim 1, configured as described above.

12. A method for assisting in the diagnosis of advanced prostate cancer, wherein the method is carried out by a system (10) including a control module (15), and the method is carried out by the control module (15) and comprises the following steps: - Receiving (110) at least one magnetic resonance image of a first portion of the peripheral region of the subject's prostate, wherein the image includes a set of pixels, The image includes, for each pixel, the apparent diffusion coefficient value of the corresponding region of the first portion. - Calculate the first score (130), - Based on the first score calculated above, it is determined that the first portion includes advanced cancer (140), Includes, The first score is calculated by the control module (15) using the following quantity x １ and quantity x ２ It is calculated as a function of at least one of the following: [Math 12] Here, W is the normalized wash-in or normalized wash-out rate of the first part, the wash-in or wash-out rate is expressed as s - 1 and estimated from the set of magnetic resonance images of the first part, and ADC is the percentile of the apparent diffusion coefficient value of the first part, mm 2 It is expressed in units of / s, a １ This is the first constant, and when W is the normalized wash-in velocity value, the first constant is -6 × 10 －３ mm ２ and -10 －３ mm ２ If it falls within the range, the percentile is 40 or less, and W is a normalized washout rate value, then the first constant is 2 × 10 －２ mm ２ and 15 x 10 －２ mm ２ It is included in between, and the percentile is 15 or less. [Number 13] Here, b １ is 10 －７ mm ２ / s ２ and 14 x 10 －７ mm ２ / s ２ A second constant that falls between the first and second parts, ADC is a percentile that falls between 20 and 45 of the apparent diffusion coefficient values ​​of all pixels corresponding to the first part, TTP is the time value to the peak of the first part, expressed in seconds, the time value to the peak is the duration from the first moment to the second moment, the first moment being the arrival time of the contrast agent in the prostate, and the second moment being the time of the maximum contrast density of the MRI signal of the first part, in the method.

13. A computer program (22) comprising a software instruction configured to carry out the method described in claim 12 when the software instruction is executed on a processor (35).

14. An information medium (50) containing a software instruction, configured to perform the method described in claim 12 when the software instruction is executed on a processor (35).

Images