WO2019037999A1

WO2019037999A1 - Method and system for simulating a human tumour

Info

Publication number: WO2019037999A1
Application number: PCT/EP2018/069936
Authority: WO
Inventors: Christoph Morhard; Sebastian Richter
Original assignee: Prokando Gmbh
Priority date: 2017-08-23
Filing date: 2018-07-23
Publication date: 2019-02-28
Also published as: DE102017119278A1

Abstract

The invention relates to a method for simulating a human tumour, comprising the following steps: determining (102) a number of tumour statuses; and selecting (104) one of the tumour statuses.

Description

Method and system for simulating a human tumor

description

The invention relates to a method and a system for simulating a human tumor.

Malignant human tumors are complex cell structures consisting, for example, of different sub-clones of a primary tumor. These sub-clones respond

different on therapies. In addition, individual treatment courses can lead to a high mutation rate and a high evolutionary pressure within the tumor and thus to a resistance to a form of therapy. This development of resistance is a reason for the failure of certain therapeutic approaches, especially in late tumor stages.

Furthermore, it is known that databases are available which include the genetic mutations of tumor cells throughout the course of the disease.

Although there are new imaging techniques and new ones

Biopsy procedure gives, it is not possible, the

Overall state of a remaining tumor in the body of the patient in particular to determine its composition with tumor cells of different genome. On

Patient-specific tumor can be considered as a black box, because the internal state is not or only partially observed. On the other hand, a large number of therapeutic options for the treatment of a malignant tumor are available.

So you could be the objective technical task

to formulate that as a method for

Simulation of a human tumor is to be created, which simulates a patient-specific development of a tumor. The problem underlying the invention is achieved by a

A method of simulating a human tumor according to claim 1 and solved by a system for simulating a human tumor according to an independent claim. According to a first aspect, a method for simulating a human tumor is proposed. This involves detecting a number of tumor status, the

Determining the respective tumor status by means of a

Simulation function is performed, and selecting one of the tumor status, wherein selecting the one

Tumor status is performed from the number of tumor statuses by means of a first selection function trained by supervised learning.

In a second aspect, a system for simulating a human tumor is proposed. This involves detecting a number of tumor status, the

Determining the respective tumor status by means of a

By doing that, first a number of tumor status

is determined, a large amount of possible tumor conditions is first provided, which is then limited by the first machine learned selection function. Consequently, a model is being created which addresses the complexity of developing a human tumor through machine learning.

According to one embodiment, the number of tumor statuses is at least one time prior to at least one determined tumor status, depending on the

Tumor development relevant biological parameters and depending on at least one at least partially lying in the past and the tumor therapy measure determined. The tumor status will depend on the number of tumor status

provided diagnostic data, depending on a patient's history and selected depending on the at least one partially lying in the past and the tumor therapy. Thus, the process or system are the essential

Information is available to deliver the best possible result.

In a particularly advantageous embodiment, a therapy proposal is determined, wherein the determination of the

Therapy proposal is performed by means of a second selection function, which is trained by reinforcing learning.

In an advantageous embodiment of the

Therapy suggestion depending on one

Treatment goal, depending on the at least one partially lying in the past and the tumor therapy measure, and determined depending on the selected tumor status. This gives the second selection function the information to

to determine the best possible result. Further advantages and features of the invention can be found in the following description of exemplary embodiments of the drawing. In the drawing: Figures la, lb, 2, 4, 5 and 6 are each a schematic block diagram;

FIG. 2 is a schematic view of a human tumor; Figures 7, 8, 9, 10, 11, 12, 13 and 14 each a

schematic flow diagram;

FIG. 15 is a schematic sequence diagram; and FIG. 16 shows by way of example a representation of

Simulation results.

Figure la shows a schematic block diagram. to

Simulation of a human tumor, a block 102 determines a number of tumor statuses Sl (t), S2 (t), SN (t) as a function of at least one temporally previously determined tumor status S (t-1), depending on the

Tumor development related biological parameters P and depending on at least one at least partially in the past and the tumor therapy T (t-l). A block 104 selects one

most likely the tumor status Sl (t), S2 (t) or SN (t) depending on provided diagnostic data D (t), depending on a patient history H and in

Dependence on the at least one partial in the

Past and tumor related

Therapy measure T (t-l) and provides this as

selected tumor status S (t) ready.

FIG. 1b shows, in addition to FIG. 1a, a schematic block diagram with a block 106, which has a

Therapy proposal R (t) for use in one

depending on a treatment goal Z (t), depending on the

at least one therapy measure T (t-1), which lies partially in the past and concerns the tumor, and in dependence on the selected tumor status S (t)

determined.

The following definitions apply to the whole of the present

Description to apply: A treatment goal in a malignant tumor disease includes, for example: complete recovery of the patient, low treatment side effects and / or long-term survival of the patient

Patients.

A therapy method relates to a single possible therapy action such as a radiological treatment or a surgical treatment or a drug treatment but not a combination of the aforementioned possible treatments.

A therapy proposal concerns a proposal that includes one or more therapeutic methods and that

within the model proposed in this description should have a positive development of the tumor status with respect to the treatment goal.

A therapeutic measure relates to a therapy actually performed on the patient, such as

For example, a radiological treatment, a surgical treatment, a drug treatment or a combination of the aforementioned treatments. For example, a patient history includes information about the patient's age, whether or not the patient was a smoker, information about whether the patient is a city dweller or not. Consequently, the patient history includes data about

Risk indicators for a particular type of cancer.

Diagnostic data D (t) at a time t include, for example, the size and / or shape of the

Patients existing tumor. This can, for example, via a computed tomography recording - too machine - be determined. Alternatively or

In addition, a tumor sample is taken from the patient at the beginning of the treatment and a cell culture

created. The tumor cells of the cell culture thus grow parallel to the patient. The diagnostic data D (t) to the

Time t thus include, for example, a size of the cell culture and / or genetic data of the

Cell culture, i. a composition

various tumor cells of the cell culture. If a therapeutic measure is performed on the patient in one example, an analogous therapy measure is applied to the cell culture.

FIG. 2 shows a schematic block diagram with a system 200 for carrying out the methods listed in this description. The system 200 includes a

central processing unit 202 such as a number of individual systems comprising data center with increased computing capacity and a remote control terminal 204 such as a commercial personal computer. The central processing unit 202

communicates over a data link 206, which is established over a wide area network, with the remote

arranged operator terminal 204th

The computing unit 202 comprises in a schematic form a first data memory 202M, a first processor unit 202P and a first network interface 2021

Operator terminal 204 includes a second

Network interface 2041, a second data store 204M, a second processor unit 204P. Further, the operator terminal 204 includes an input interface 204T such as a keyboard and / or a keyboard

Computer mouse and an output interface 204S like

for example, a screen. Input interface 204T and output interface 204S are collectively referred to as user interface I.

FIG. 3 shows by way of example a malignant human tumor in the form of a tumor status S in a schematic manner

View. The tumor status S comprises a number of type A cells Ca, a number of type B cells Cb, and a number of types C cells. In addition to malignant tumors, benign tumors can also be used depending on the patient and

Disease history of interest.

FIG. 4 shows a schematic block diagram of a first model. A simulation function SIM determines the tumor status Sl (t), S2 (t), SN (t). The simulation function SIM simulates several times the growth of a tumor, for example, based on a past tumor status Sl (t-1) at time t-1. Of course, several tumor statuses may also be available on the basis of which the simulation function SIM determines a number of different tumor statuses. The simulation function SIM

Rule-based determines a number of tumor states, with the rules defining a respective one

Include transition probability from one cell status to the next cell status. In one embodiment of the simulation function SIM, diagnostic data of the respective patient are provided

considered that a respective fully or partially simulated tumor status is not provided as a result. If z. B. a Biospieerrgebnis is present, from which it follows that a certain mutation is present, in the simulation function u. U. rejected large numbers of detected branches in which this particular mutation is not present.

The most recently applied therapeutic measure T (t-l), which has been carried out on the patient up to the present time, is stored in a patient-individual memory area M2 and is available for retrieval with further therapy measures that are further in time. Presently available diagnostic data D (t), which relate to the actual tumor of a patient, are stored in a patient-specific memory area M3

are stored and ready for retrieval with further diagnostic data further back in time. A

Patient history H relates to the condition of the patient before the occurrence of the tumor disease and is stored in a patient-specific memory area M4. Based on the tumor status Sl (t-l), derived from a database DB biological parameters P and at least one lagging ago and the tumor concerned

Therapy measure T (tl) is a number of tumor status Sl (t), S2 (t) at time t and a number of tumor status Sl (t + 1), S2 (t + 1), S3 (t + 1 ) and S4 (t + 1) at the time t + 1 determined. Assuming the time t as the present and the time t + 1 as the future, starting from a number of current estimated tumor statuses Sl (t), S2 (t), a number of future estimated tumor status Sl (t +1) to S4 (t + 1) closed.

The tumor statuses Sl (t) to S4 (t + 1) are called tree

represented and can include a number of trees so a forest. This forest in the form of simulation data SD is patient-individual and is stored in the patient ^¬ individual memory area Ml.

A first selection function SEL1 is based on machine learning and, for example, supervised learning

trained. The first selection function SELl provides each

Edge of the tree with an associated probability, resulting in a most likely development course V of the actual tumor. So the first selection function SELl chooses one - from the point of view of the first one

Selection function SELl - most likely

Development course V out. Consequently, the chooses

Selection function SEL1 also the tumor status S (t + 1) from the number of determined tumor status Sl (t + 1) to S4 (t + 1), which comes closest to a state of the actual tumor in the future. To select the

Development process V the selection function SELl the simulation data SD, the patient history H and the diagnostic data D (t) are provided. The most likely course of development V of the simulated tumor will help a treating physician to decide on the next Treatment of the patient brought to the knowledge. By means of the development course V provided, the physician can initiate a self-selected therapeutic measure for the patient.

In a form not shown here, for example, a comparison of the current diagnosis data D (t) with the most probable tumor status Sl (t) can be brought to the attention of the attending physician. In this case, for example, the size of the actual tumor according to the diagnostic data D (t) with the size of the simulated tumor according to the tumor status Sl (t) are displayed. This display can be done, for example, by a figurative representation or by an estimate error determined from the comparison in the form of a number. Reference is made to the exemplary representation in FIG. 16.

In one embodiment of the first selection function SEL1, diagnostic data D (t) of the respective patient

is taken into account as the provided simulated tumor status Sl (t) to S4 (t + 1) is not selected. If z. B. is a Biospiergebnis, it follows that a certain mutation is present

at least that tumor status Sl (t) to S4 (t + 1) is not selected, in which this particular mutation is not

present.

FIG. 5 shows a schematic block diagram of a second model. For the matching with Figure 4 features reference is made to the description there. On Treatment goal Z (t) is given, can not in

be stored and thus is also in the future for future calculations to

Available.

A second selection function SEL2 is based on machine learning and is trained, for example, by reinforcing learning. The second selection function SEL2 determines the therapy proposal R (t), which includes one or a number of therapy methods Ma, Mb, Mc, to Mx. The

Therapy methods Ma to Mx are stored in another database DB2. The second selection function SEL2 selects from the further database the therapy methods Ma to Mx, which depend on the previous ones

Therapy measures T (t-l), the course of development V and the patient history H the therapy goal Z (t) best

correspond. The selection functions SEL1 and SEL2 are, according to this description, for example, machine-trained neural networks.

FIG. 6 shows a schematic block diagram of a third model. For the matching with the figures 4 and 5 features reference is made to the description there. In contrast to FIG. 5, the third model of FIG. 6 does not include a first selection function SEL1. The possible

Development course V of the simulation function SIM is selected, for example, at random. The second selection function SEL2 selects from the other

Database the therapy methods Ma to Mx, which in response to the previous therapeutic measures T (t-l), the development course V and the patient history H the therapy goal Z (t) best.

FIG. 7 shows a schematic flow diagram for

Initialization of the simulation function SIM from FIGS. 4-6. In this description, a cell is to be understood as a data structure which is a biological structure

Cancer cell and its environment, such as

Blood vessels, mimicking or their properties in particular has their genetic information. In a step 702, an initial cell is determined as a function of stochastic mutation probabilities and as a function of the patient's history and added to a list. In a step 704, the next cell of the list is selected for editing. In a step 706

decided whether predetermined initialization parameters such as the predetermined tumor size have already been reached. If so, at step 708, the

stored list stored as an initial tumor status. If this is not the case, then a change is made to a step 900, which is described in greater detail with respect to FIG. 9.

Figure 8 shows a schematic flow diagram of

Simulation function SIM for existing tumor status in the form of a cell list. In a step 802, the cell list with a number of cells

provided, wherein the provided cell list corresponds to the previously determined tumor status. In a step 804, the next cell is selected from the list. A step 806 checks whether all cells from the list have already been processed in this simulation step. If this is the case, then in step 808 the determined list is stored as a further tumor status. If this is not the case, it is changed to step 900, which is described in more detail in FIG. 9.

Figure 9 shows a schematic flow diagram for

Operating the simulation function SIM. In a step 904 it is probabilistically determined whether a first cell

dies. For this a cell environment and the

Zellfitness determined and as a result, whether the first cell dies. For this purpose, first an individual cell fitness is determined and depending on the

determined cell fitness then the mortality. In step 908, first mutation parameters are applied to the first cell when the first cell is not dead. Thus, in step 908, a

Change of the genome made, which not through a cell division is caused. In a step 912 it is probabilistically determined if the first cell splits if the first cell is not dead. In a step 914, a second cell is determined in dependence on the state of the first cell and added to the list

added when the first cell has split. In a step 916, second mutation parameters are applied to the first cell and / or the second cell when the first cell has split. Thus, in step 916, a change of the genome of the first and / or the second cell is effected, which is caused by a cell division. The foregoing mutation parameters are determined, for example, as a function of the biological parameters P and as a function of the therapeutic measure T (t-1).

FIG. 10 shows a schematic flowchart which represents the procedure for generating a therapy proposal. In a step 1002, at time t

present diagnostic data and the time t

present medical history of the patient comprising the patient history and the previously used

Therapeutic measures provided. In a step 1004, the diagnostic data is processed, for example, a tumor size and / or a tumor shape becomes

Computer tomography images. Optionally, step 1004 may also be replaced by a manual entry of the physician. In a step 1006, the

Simulation function SIM performed from the previous figures to a number of different tumor status too determine. In a step 1008, the first selection function SEL1 from the previous figures is executed to select the tumor status that best suits the

specific patient fits. In a step 1010, the second selection function SEL2 from the previous figures is executed to select the therapy proposal that best suits the specific patient. In one

Step 1012 visualizes the selected tumor status and the selected therapy proposal. In one

Step 1014 checks whether new diagnostic data is available at a subsequent time. If so, step 1002 is entered. If this is not the case, then a change is made to a step 1200, which is described below in FIG. Alternatively, instead of step 1200, step 1100 is performed

changed, which in the following figure 11

is shown.

FIG. 11 shows a schematic flow diagram for

Manual prescription of a therapy proposal to a

Development under the application of the given

Therapy suggestion to simulate. In step 1102, a therapy proposal is manually provided. In a step 1104, the simulation function SIM from one of the previous figures is performed and a number of tumor statuses are determined. In a step 1106, one of the previously determined tumor status is selected by means of the first selection function SEL1. In a step 1108, the selected tumor status that was determined for the manually prescribed therapy proposal is visualized. In one Step 1110 checks if there is another one

Therapy suggestion should be specified manually. If so, step 1102 is entered. If this is not the case, then a change is made to step 1112 and the method is ended.

FIG. 12 shows a schematic flow diagram for

Generation of a therapy proposal without application of the first selection function SEL1. In a step 1210, the simulation function SIM executed is determined as a number of tumor statuses as a function of a previously determined tumor status. In a step 1212, by means of the second selection function SEL2, at least one of the

determined tumor status selected a therapy proposal. In a step 1214, the tumor status is updated with the

associated therapy proposal at time t

visualized. In a step 1216, it is checked if another pass is required. If this is the case, then the step 1210 is changed. If this is not the case, then a change is made to step 1218 and the process is ended.

FIG. 13 shows a schematic flow chart for the training of the first selection function SEL1 through supervised learning. In a step 1304, an actual

Course of a tumor disease comprising an actual first tumor status and an actual second tumor status SR (t) from a disease database. In a step 1308, a number of simulated tumor status is obtained from the first actual tumor status determined. In a step 1310, one of the determined simulated tumor status S (t) is selected by means of the first selection function SEL1. In a step 1312, the simulated tumor status S (t) having the second actual tumor status SR (t)

compared to determine an estimation error. In a step 1318, the first selection function SEL1 in FIG

Updated dependence on the estimated error detected. The selection function SEL1 is trained by the estimation error.

FIG. 14 shows a schematic flow chart for the training of the second selection function SEL2

encouraging learning. In a step 1402, a third tumor status S (t) is determined. In a step 1404, a therapy proposal R (t) is determined as a function of the third tumor status S (t). In a step 1406, a fourth tumor status S (t + 1) is determined as a function of the third tumor status S (t) and in dependence on the selected therapy proposal R (t). In step 1408, a reward value is determined based on a reward function in response to the fourth tumor status S (t + 1). In a step 1416, the second selection function SEL2 becomes dependent on the determined reward value

updated. The reward value is determined as a function of a predetermined treatment goal Z (t). This means that both the training and the operation of the second selection function has several modes depending on the treatment goal Z (t). The second selection function SEL2 is trained using the reward value. FIG. 15 shows a schematic sequence diagram. A physician A examines a patient P and uses the acquired information to determine patient history H as well

Diagnostic data D (t). The patient history H as well

Diagnostic data D (t) is provided by the physician via the

Input user interface I and are thus the simulation function SIM and the two selection functions SEL1 and SEL2 available. The patient history H and diagnostic data D (t) are used to initialize the

Transfer simulation function SIM. After initialization, the simulation function requires only the therapeutic measures T and the diagnostic data D (t) performed on the patient P. In a step 1502, the simulation function SIM is executed and determines tumor status Sn. The tumor statuses Sn are used by the first selection function SEL1 in a step 1504 to provide one of the delivered tumor statuses Sn as the selected tumor status Ss of the second selection function. The second

Selection function SEL2 determines in step 1506

Therapy proposal R (t). The selected tumor status Ss (t) and the determined therapy proposal R (t) become the

User interface I provided. The

User interface I determines a visualization VIS of the tumor status Ss (t) and the determined mediated therapy proposal R (t), which is made available to the doctor A.

Depending on the data provided, the physician A can now decide which therapy measure T (t) he selects P for the patient. The therapy measure T (t) may coincide with the therapy proposal R (t) or not. The doctor A transmits the applied

Therapy measure T (t) to the user interface I, with which it is available to the entire system.

In the embodiment of the system 200 of FIG. 2

the central processing unit 202 determines the number of tumor statuses Sl (t), S2 (t), SN (t) by means of the

Simulation function SIM. The central processing unit 202 transmits the ascertained tumor statuses Sl (t), S2 (t), SN (t) to the operating terminal 204. The remotely located operator terminal 204 selects one of the simulated tumor statuses Sl (t), S2 (t) , SN (t) by means of the first selection function SEL1. The remotely located operator terminal 204 determines the therapy proposal R (t) by means of the second

Selection function SEL2. The remotely located

Operator terminal 204 displays the determined tumor status S (t) and the determined therapy proposal R (t) by means of the user interface I.

The central processing unit 202 is designed to train the first and / or the second selection function SEL1, SEL2. After completion of the respective training, the central processing unit 202 transmits the trained first and / or second selection function SEL1; SEL2 to the operator terminal 204. The operator terminal 204 replaces a previously used first and / or second selection function SEL1; SEL2 by the transmitted first and / or second selection function SEL1; SEL2. Of course, the simulation function SIM, the two selection function SEL 1 and SEL 2 and the

User interface I in different

Embodiments 200, 1520, 1530 may be distributed differently to the central processing unit 202 and the operator terminal 204.

FIG. 16 shows by way of example a representation of

Simulation results obtained according to the methods used here.

Claims

claims

A method of simulating a human

Tumor, the method comprising:

- determining (102) a number of tumor statuses, wherein the determination (102) of the respective tumor status is performed by means of a simulation function (SIM); and

Selecting (104) one of the tumor statuses, wherein selecting (104) the one tumor status from the number of tumor status is performed by a first selection function (SEL1) trained by supervised learning.

The method of claim 1 comprising:

- Determine (102) the number of tumor status in

Dependence on at least one previously determined tumor status, depending on the tumor development biological parameters in question and depending on at least one at least partially in the past and the tumor therapy in question; and

Selecting (104) the most probable tumor status from the number of tumor statuses depending on provided diagnostic data, depending on a patient history and depending on the at least partially in the past

lying and the tumor-related therapy. The method of claim 1 or 2 comprising:

- Determining (106) a therapy proposal, wherein the determination (106) of the therapy proposal (R) by means of a second selection function (SEL2) is performed, which is trained by fortärkendes learning.

The method of claim 3 comprising:

Determine (106) a therapy proposal in

Dependence on a treatment goal, in

Dependence on the at least one partially lying in the past and the tumor therapy measure, and depending on the

selected tumor status.

The method of any one of the preceding claims, wherein the simulation function (SIM) comprises:

- providing (802) a cell list with a number of cells, wherein the provided cell list corresponds to the previously determined tumor status;

- determining (904) if a first cell dies;

Apply (908) first mutation parameters to the first cell if the first cell is not

died;

- determining (912) if the first cell divides if the first cell is not dead;

- adding (914) a second cell to the list when the first cell has split; and

Apply (916) second mutation parameters the first cell and / or the second cell when the first cell has split.

The method of any one of the preceding claims, wherein the training of the first selection function (SEL1) by supervised learning comprises:

- providing (1304) an actual history of a tumor disease comprising an actual first tumor status and an actual second tumor status;

Determining (1308) a number of simulated tumor status from the first actual tumor status;

- selecting (1310) one of the determined simulated tumor statuses;

- comparing (1312) the simulated tumor status with the second actual tumor status to determine an estimation error; and

Updating (1318) the first selection function (SEL1) in dependence on the determined one

Estimation error.

The method of claim 6, wherein the training of the second selection function (SEL2) by reinforcing learning comprises:

- determining (1402) a third tumor status;

Determine (1404) a therapy proposal in

Dependence on the third tumor status;

Determining (1406) a fourth tumor status as a function of the third tumor status and in Dependence on the selected therapy proposal;

- determining (1408) a reward value in

Dependence on the fourth tumor status;

- updating (1416) the second selection function (SEL2) in dependence on the determined one

Reward value.

A computer program product adapted to carry out the method of any one of the preceding claims. A system (200) for simulating a human tumor, wherein the system (200) is configured to:

- To determine a number of tumor status, wherein the determination of the respective tumor status by means of a simulation function (SIM) is feasible; and

selecting one of the tumor statuses, wherein selecting the one tumor status from the number of tumor statuses is feasible by means of a first selection function (SEL1) trained by supervised learning.

The system (200) according to claim 9, which is adapted to the method according to one of

Claims 2 to 7 execute.