WO2014096147A2 - Détermination d'une valeur d'indicateur de probabilité - Google Patents

Détermination d'une valeur d'indicateur de probabilité Download PDF

Info

Publication number
WO2014096147A2
WO2014096147A2 PCT/EP2013/077309 EP2013077309W WO2014096147A2 WO 2014096147 A2 WO2014096147 A2 WO 2014096147A2 EP 2013077309 W EP2013077309 W EP 2013077309W WO 2014096147 A2 WO2014096147 A2 WO 2014096147A2
Authority
WO
WIPO (PCT)
Prior art keywords
phenotypical
features
image data
illness
input image
Prior art date
Application number
PCT/EP2013/077309
Other languages
English (en)
Other versions
WO2014096147A3 (fr
Inventor
Maria Jimena Costa
Andre Reis
Michael Sühling
Georgia VASILEIOU
Original Assignee
Siemens Aktiengesellschaft
Friedrich-Alexander-Universität Erlangen-Nürnberg
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Siemens Aktiengesellschaft, Friedrich-Alexander-Universität Erlangen-Nürnberg filed Critical Siemens Aktiengesellschaft
Priority to EP13819022.8A priority Critical patent/EP2901340A2/fr
Publication of WO2014096147A2 publication Critical patent/WO2014096147A2/fr
Publication of WO2014096147A3 publication Critical patent/WO2014096147A3/fr

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/0002—Inspection of images, e.g. flaw detection
    • G06T7/0012—Biomedical image inspection
    • G06T7/0014—Biomedical image inspection using an image reference approach
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
    • G16B20/00—ICT specially adapted for functional genomics or proteomics, e.g. genotype-phenotype associations
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10072—Tomographic images
    • G06T2207/10088—Magnetic resonance imaging [MRI]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/20—Special algorithmic details
    • G06T2207/20076—Probabilistic image processing
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30004—Biomedical image processing
    • G06T2207/30068—Mammography; Breast
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30004—Biomedical image processing
    • G06T2207/30096—Tumor; Lesion

Definitions

  • the present invention concerns a method of determining a probability indicator value representing a probability, i.e. a likelihood, of a presence of a genetic predisposition of a being for a specific illness, in particular of a particular organ or structure.
  • the invention also concerns a determina- tion assembly for such purpose.
  • the invention refers to probability indicators in the field of cancer research, but it may also be applied for predispositions for other illnesses. Therefore, reference is mainly made throughout this description to cancer and in particular to breast cancer, fully bearing in mind at any time the much broader possibility of use of the invention.
  • BRCA1 breast cancer suscep- tibility gene 1
  • BRCA2 breast cancer susceptibility gene 2
  • Hereditary breast cancer is associated with a poorer sur- vival, and hereditary ovarian cancer with a better survival than their sporadic counterparts, an observation which can affect the treatment strategies, (cf. van der Groep, P. et al . : Distinction between hereditary and sporadic breast cancer on the basis of clinicopathological data. J Clin Pathol 2006; 59, pp. 611-617) .
  • the surveillance strategies that are suggested for the carri ⁇ ers of the mutations include more frequent and detailed imag ⁇ ing screening in respect to sporadic cancer patients (cf. Daly, Mary B. et al . : Genetic/Familial High-Risk Assessment: Breast and Ovarian. Journal of the National Comprehensive Cancer Network. 2010;8:5, pp. 562-594).
  • the car ⁇ riers are at high risk of developing other forms of cancer such as early-onset prostate cancers and pancreas cancer.
  • the early detection of the mutations may affect strategies for clinical management of these cancers (cf. Diamandis, Maria et al . : Personalized Medicine: Marking a New Epoch in Cancer Pa ⁇ tient Management. Mol Cancer Res 2010; 8:9, pp. 1175-1187).
  • the method mentioned in the introductory paragraph thus comprises the following steps according to the invention: a) In a first step, a number of input image data of an organ or structure is provided which potentially shows phenomena of the specific illness from a tomographic imaging device.
  • Such tomographic imaging device thus comprises a technical apparatus realized to automatically (upon initiation by a user) acquire images of the inner body of the being.
  • Such to ⁇ mographic devices comprise for instance mammography devices, computer tomographs (CT) , magnetic resonance systems (MR) , ultrasound systems, X-ray devices, angiographs, single photone emission computer tomographs (SPECT) , positron emis- sion tomographs (PET) and many others.
  • the image data taken by that tomographic imaging device com ⁇ prise image data of a suspicious organ or structure, i. e. of the target within the body of the being which might be af- fected by the specific illness and which thus might show phe ⁇ nomena of the illness.
  • Such phenomenon may for instance be a tumor if the illness is cancer but may comprise in fact any phenomenon which can be connected to the specific illness.
  • "illness” is defined to be the overall diagnosis (in this case cancer)
  • the phenomenon may be any physical manifestation of such illness (such as the tumor in this case) but also other physical side-effects of the illness which can be connected to the illness.
  • the input image data are analyzed by ex ⁇ tracting therefrom a number of low-level phenotypical fea ⁇ tures .
  • the low-level phenotypical features are ex- tracted from the image data based purely on the image data with no additional input from other sources. In other words, nothing else is used for extraction of the low-level pheno ⁇ typical features than only the input image data and the ex- traction method in a (physical and/or software-based) extrac ⁇ tion system.
  • these low-level features are preferably re ⁇ lated directly to the specific illness. Possible relations between a specific illness and phenotypical features will be indicated below for the case of breast cancer, but it may be noted that every specific illness can potentially relate to specific phenotypical features in image data which specific phenotypical features certainly vary from illness to illness.
  • a "phenotypical feature” represents a phenotypical phenomenon of the organ and/or structure in question. That means that the phenotypical feature may directly characterize the pheno ⁇ typical phenomenon or it may be such feature that is derived from that phenotypical phenomenon.
  • the definition of "phenotype” is generally very broad as it comprises all observable characteristics or traits of an or ⁇ ganism such as morphology, development, biochemical and physiological and anatomical properties, phenology and even behaviour. In the context of the invention, the focus lies clearly on morphology, development, physiology and anatomy, very particularly on microscopic morphology (shapes) and anatomy, i. e. the microscopic structures (texture, density etc.) of the organ or structure in question.
  • a third step there is provided a plurality of pheno ⁇ typical reference features derived from reference image data of the same organ or structure of a number of reference be ⁇ ings and/or a database with a number of datasets representing the phenotypical reference features, the plurality of pheno- typical reference features comprising such phenotypical ref ⁇ erence features which correspond with the phenotypical fea ⁇ tures of the input image data.
  • the focus is on phenotypical reference fea ⁇ tures: a plurality of such phenotypical reference features, preferably a plurality of sets of such phenotypical reference features is used directly or indirectly (via the number of datasets) .
  • the phenotypical reference features correspond with the phe ⁇ notypical features of the input image data. That means that either the type of phenotypical features of the input image data is chosen dependent on the type of phenotypical refer ⁇ ence features or vice versa in such way that the former and the latter comprise features which can be compared in such way that a comparison can possibly lead to an indicative re- suit. Most preferred, the former and the latter features are based on the same parameters and/or underlying questions.
  • phenotypical reference features themselves or of datasets representing them.
  • datasets may thus comprise an extraction and/or interpretation and/or translation of data from the phenotypical reference features into a different data system, for instance from parameter values into vectors or the like.
  • the phenotypical features of the input image data are processed through a classification and/or comparison algorithm based on the number of phenotypical reference features and/or on the datasets.
  • the phenotypical reference features and the phenotypical features of the input image data are brought to ⁇ gether directly or indirectly (via a classifier) in order to find out to which degree the phenotypical features of the in- put image data match with certain ones of the phenotypical reference features.
  • the comparison and/or classification of the phenotypical features of the input image data based on the phenotypical reference features is used to generate the probability indi ⁇ cator value as a final result.
  • the method according to the invention is most preferred auto ⁇ matic or at least partially automatic such that at least steps b) , d) and e) are fully automatic, i.e. without any hu ⁇ man influence.
  • the method according to the invention is based on two new findings and assumptions by the inventors, which findings are based on cancer phenomena but which can be extended to other illnesses as well.
  • a first assumption is that certain pheno- typical features on the microscopic level are specifically influenced by illnesses. This has been proven for the first time by tests of the inventors in the case of breast cancers induced due to BRCA1 or BRCA2.
  • a second finding by the inventors was that even although such changes of phenotype may not be visible to the naked eye in tomographic image data - even for the experts - they can nevertheless be extracted there ⁇ from, for instance by using algorithm-based approaches. Some particular approaches of that kind will be explained in de ⁇ tail below.
  • the invention is therefore based on low-level (phenotypical ) features of a patient, i.e. such features which can be de ⁇ rived exclusively from image data as opposed to high-level features such as the ones which will be mentioned below.
  • the invention uses for the first time highly reli- able (and essentially reproducable) evaluation material in the form of the image data from a tomographic imaging device for assessment of types of illnesses rather than (in the case of cancer research) rather vague information such as family background etc.
  • the probability indicator values which can be determined that way can thus be more accurate and precise whilst they can be acquired even by using existing such image data or by using image data which need to be acquired anyway in the course of evaluation or treatment if the specific ill ⁇ ness in question.
  • a determination assembly as mentioned in the introductory paragraph according to the invetion comprises: a) A first provision unit which in operation provides a num- ber of input image data from a tomographic imaging device of an organ or structure which potentially shows phenomena of the specific illness. Such provision unit is realized to per ⁇ form step a) of the method according to the invention. It may comprise or be directly connected to an acquisition unit re- alized to acquire the input image data and/or it may be real ⁇ ized as an input interface to receive such input image data from such acquisition unit and/or from a memory in which these input image data are stored after their acquisition. b) An analysis unit for analyzing the input image data by ex ⁇ tracting therefrom a number of low-level phenotypical fea ⁇ tures.
  • the analysis unit is thus realized to perform step b) of the method according to the invention.
  • a second provision unit which in operation provides a plurality of phenotypical reference features derived from refer ⁇ ence image data of the same organ or structure of a number of reference beings, the plurality of phenotypical reference features comprising such phenotypical reference features which correspond with the phenotypical features of the input image data.
  • the second provision unit thus performs step c) of the method according to the invention. It may comprise or be directly connected to a database in which the plurality of phenotypical reference features and/or the datasets are stored and/or generated and/or it may be realized as an input interface to receive these phenotypical reference features and/or the datasets.
  • a derivement unit for deriving the probability indicator value from the processing That derivement unit is thus real ⁇ ized to perform step e) of the method according to the inven- tion. It may be comprised by the processing unit or realized as an independent operating unit.
  • the probability indicator value may for instance comprise a confidence factor in the form of a percentage value, but other possibilities include, for instance, a ranked list of similar cases on which statis- tics about genetic predisposition could be computed.
  • the invention also concerns a tomographic imaging device with an acquisition unit and a determination assembly according to the invention.
  • the acquisition unit can thus be used to ac- quire, i. e. generate the input image data of the being.
  • the imaging device according to the invention, and/or the determination assembly according to the invention, in particu- lar its first and/or second provision units, the analysis unit, the processing unit and the derivement unit (but also other components of the determination which are mentioned be ⁇ low) may be partially or wholly accomplished by hardware com- ponents, for example using semiconductor chips such as ASICs (Application Specific Integrated Circuits) , FPGAs (Field Pro ⁇ grammable Gate Arrays) , or PLAs (Programmable Logic Arrays) . They may, however, also be comprised of software components or combinations of hardware and software components.
  • ASICs Application Specific Integrated Circuits
  • FPGAs Field Pro ⁇ grammable Gate Arrays
  • PLAs Programmable Logic Arrays
  • the invention also concerns a computer programme prod ⁇ uct computer programme product directly loadable into a proc ⁇ essor of a programmable determination assembly comprising programme code means to conduct all steps of a method accord ⁇ ing to the invention when the computer programme product is executed on the determination assembly.
  • the method according to the invention can principally also be applied in the case of illnesses of plants, it is preferred that the above-mentioned being is a human being or an animal, most preferred a living human being or animal. A particular focus is certainly on human beings as becomes clear from the reference example of hereditary breast cancer evaluation .
  • the organ or structure is a breast and/or an ovary and/or a prostate and/or a colon and/or a fallopian tube of a human being or of an animal.
  • all these organs or structures can potentially be af ⁇ fected by hereditary breast cancers due to BRCA1 and BRCA2.
  • the specific illness comprises a cancer, most preferred a specific cancer type (such as breast cancer or hereditary breast cancer) .
  • the method is particularly preferred in the context of the genetic predisposition comprising the presence of a breast cancer suceptility gene, preferably the breast cancer susceptility gene 1 and/or the breast cancer susceptility gene 2.
  • Other tumor-developing illnesses due to genetic predispositions comprise the Li-Fraumeni syndrome and the Cowden syndrome (cf. Daly et al . , see above) .
  • other specific ill ⁇ nesses due to genetic predispositions also in other fields than cancer can be evaluated by the determination method ac ⁇ cording to the invention.
  • degenerative ataxias such as morbus Friedreich (Friedreich's ataxia) or in yet another field or hereditary disabilities of patient's to react to beta blockers in a de- sired way.
  • a particular abnor ⁇ mality region is selected and used in the context of steps b) to e) .
  • Such abnormality region also called region of inter- est -comprises an abnormality due to the illness under con ⁇ sideration, for instance comprises a tumor (or another side- effect of cancer) in case of a cancer such as (potentially hereditary) breast cancer. Selecting such abnormality region has the effect that one can focus much clearer on a smaller part within the input image data so that the necessary compu ⁇ tational power to carry out the steps of the invention can be reduced which makes the overall process faster and also more reliable .
  • the step d) comprises a classification by means of a trained classifier algorithm.
  • a trained classifier algorithm may for instance comprise a Probabilistic Boosting Tree algorithm.
  • the trained classifier algorithm comprises a Random Forest clas ⁇ sifier .
  • Random Forest comprises a set of k decision trees
  • each tree provides leaves, i.e. terminal positions or terminal nodes of the tree (also called “terminals”) , indicating a classification or regression range based on the query instance.
  • the trained Random Forest provides a model for a specific categorization or regression problem, which categorization or regression problem is in the following called “context”.
  • the "model out ⁇ put” or “model” yield for a specific context is based on the leaves given by the Random Forest for a respective instance (i.e. set of features as defined for example for a query in ⁇ stance or other instances) propagated down all k trees.
  • the vote or yield of the Random Forest i.e.
  • a categorization or regression range determined is given by the category or re ⁇ gression range indicated by the majority of leaves given by the Random Forest for a specific instance, respectively the query instance.
  • the query instance in the context of the in ⁇ vention is defined by the phenotypical features of the input image data which are processed through the Random Forest. It should be noted that an option for providing a trained
  • Random Forest is the training of a Random Forest by incorpo ⁇ rating "cases" or data, i.e. images. Incorporating cases or data to train a Random Forest is referenced in Breiman L., Random Forests, Machine Learning 2001; 45:1, pp 5-32. In the context of the method according to the invention, such
  • cases or data comprise the phenotypical reference features which can be used to train the random forest.
  • the phe- notypical features of a current case may be used to further train the Random Forest (as is indeed the case for other trained classifier algorithms as well) . Therefore, trained classifier algorithms such as the Random Forest can be em- ployed as self-learning algorithms with an ever-increasing precision of determination of probability indicator values.
  • phenoypical features used in the context of the invention, these can be divided into at least two categories, namely a first group of such features which can be derived directly from the input image data and a second group of such features which need to be processed through a more refined evaluation algorithm, i.e. derived indirectly via an inter- pretation and/or interpolation and/or recognition engine
  • the first group of phenotypical features comprises for in ⁇ stance
  • an intensity i.e. a density (for instance measured in Hounsfield Units - HU)
  • the second group of phenotypical features comprises for in- stance
  • - results from a relative frequency histogram and/or of up to four first central normalized moments thereon for a desnsity (e.g. HU) distribution (in a bounding box of a selected part of the organ/structure, e.g. of a tumor or lesion),
  • desnsity e.g. HU
  • a bounding box of a selected part of the organ/structure e.g. of a tumor or lesion
  • image moment invariants may for instance be image moment in ⁇ variants by Hu (as described in Hu, Ming-Kuei: Visual Pattern Recognition by Moment Invariants. IRE Transactions on Infor ⁇ mation Theory 1962, pp. 179-187) and/or by Zernike (as de- scribed in Pejnovic, P. et al . : Object Recognition by Invari ⁇ ants. Proc Int Conf on Pattern Recognition 1992, pp. 434- 437) .
  • phenotypical features of the second group has been proven particularly useful in the context of the invention as they can easily be processed through a trained classifier algorithm and as their strength of asses- ment goes far beyond an assessment of features which could potentially also be detected by the naked eye at a sufficient resolution of the input image data on a means of display. Test of the use of any of these phenotypical features in the context of the determination of probability indicator values relating to BRCA1 and BRCA2 with the method according to the invention have shown quite accurate probability results.
  • the method according to the invention can be further enhanced by using in step d) additional data, in particular high-level features or data, referring to the being.
  • additional data in particular high-level features or data, referring to the being.
  • Such high-level features such as semantic information and/or other informa ⁇ tion concerning the being from a medical practitioner can be introduced (for instance again into a trained classifier al ⁇ gorithm) into the method in order to further refine it.
  • those criteria which have been used according to the above-mentioned extensive prior art can also be used ad ⁇ ditionally in the context of the invention.
  • the additional data may be divided into at least two principal categories or groups.
  • the first group comprises those additional data which comprise data related to the be ⁇ ing, in particular data representing at least one of:
  • the second group comprises those additional data which com ⁇ prise background data of the being, in particular data repre- senting at least one of:
  • the inventors incorporate imaging evidence from image data of a tomographic imaging de ⁇ vice for the assessment of the likelihood that for instance tumors are related to a genetic mutation, as opposed to spo ⁇ radic cases of cancer.
  • Individuals with a high probability of having a genetically predisposed cancer could be referred to genetic counseling quickly and early on, allowing for specific surveillance, prevention and treatment options, avoid ⁇ ing unnecessary costs as well as potential harm resulting from the application of inadequate and potentially harmful treatment measures not adapted to genetic mutation carriers.
  • the method according to the invention can be used for deci ⁇ sion support, e.g. as complementary screening criteria (in addition to family history) , in order to refer patients to genetic counseling promptly, avoiding unnecessary costs and treatments not adapted to mutation carrier cases.
  • the method incorporates imaging (i.e. imaging phenotype) evi ⁇ dence, and possibly morphological and biological evidence, to assess the likelihood that a given tumor is linked to a ge- netic predisposition (e.g. genetic mutation).
  • imaging i.e. imaging phenotype
  • morphological and biological evidence to assess the likelihood that a given tumor is linked to a ge- netic predisposition (e.g. genetic mutation).
  • This method could be used, for instance, as a complement to current screening criteria for patient referral to genetic counsel ⁇ ing, in order to focus on individuals most likely to prove positive and offer more tailored patient management in all cases.
  • the method can be applied, for instance, to support clinicians during breast cancer screening, providing a pheno- type-based criterion for genetic susceptibility.
  • the crite ⁇ rion would support decisions about earlier referral to ge ⁇ netic counseling of those individuals most likely to prove mutation carrier positive, and leading to the personalized, adequate treatment of patients who become
  • Fig. 1 shows a schematic block diagramme of an embodiment of the method according to the invention
  • Fig. 2 shows an example of input image data which can be used in the context of the method according to the invention
  • Fig. 3 shows a first example of phenotypical features which can be derived from the input image data of Fig. 2
  • Fig. 4 shows a second example of phenotypical features which can be derived from the input image data of Fig. 2,
  • Fig. 5 shows a third example of phenotypical features which can be derived from the input image data of Fig. 2,
  • Fig. 6 shows a schematic block diagramme of a tomographic im- aging device according to an embodiment of the invention.
  • Fig. 1 shows a schematic block diagramme of an embodiment of the method Z according to the invention, details of which will be explained along Figs. 2 to 5.
  • the method Z comprises the following steps:
  • Y input image data ID - in this case from an MR device - are provided. These may be received from the MR device directly or from a memory which stores them intermedi ⁇ ately, for instance from a PACS system in a hospital.
  • the in ⁇ put image data ID comprise image data of an organ or struc ⁇ ture which potentially shows phenomena of the specific ill- ness which is to be evaluated in the course of the method ac ⁇ cording to the invention.
  • the illness in question is breast cancer
  • the structure of which the in ⁇ put image data ID are taken is a female breast 1 with a breast tumor 3 (cf. Fig. 2)
  • the probability indicator value which is to result from the method refers to the prob ⁇ ability of the person whose breast is screened being a car ⁇ rier of either BRCA1 or BRCA2.
  • a region of interest ROI - which can also be la- belled an abnormality region ROI - which contains those image data which represent the area of the (target) breast tumor 3 is selected automatically, e.g. by a detection system, and/or manually by a user (such as a medical practitioner) .
  • a user such as a medical practitioner
  • the input image data ID are analysed.
  • low-level phenotypical features PF such as the ones mentioned above are extracted from the input image data ID.
  • Those low-level phenotypical features PF comprise for in ⁇ stance a relative frequency histogram PFi (cf. Fig. 3) and four first central normalized moments on it for a density distribution in the bounding box for the lesion, i.e. the breast tumor 3, first normalized central moments PF 2 on a vector of density distribution in the bounding box of the breast tumor 3, and image moment invariants of Hu or Zernike, the latter of which is depicted as phenotypical features PF 3 in Fig. 5.
  • these depictions show different distinctive grey- scale patterns with a clear direction and/or shape. These patterns can serve to be compared with corresponding patterns (based on the same chosen Zernike parameter values) of pheno ⁇ typical reference features, as will be described below.
  • a third step W there is provided a plurality of corre ⁇ sponding (with the phenotypical features PF, PFi, PF 2 , PF 3 ) phenotypical reference features PRF derived from reference image data of other female breast of a number of other female persons.
  • a database with a number of datasets representing these pheno ⁇ typical reference features PRF.
  • a fourth, optional step V comprises the provision of addi- tional information, i.e. additional data AI such as data re ⁇ ferring to the person's history (such as previous cases of ovarian and/or breast cancer of the person), the person's risk level (based on family history, in particular referring to previous cases of breast cancer in the person's family, ), the age of onset of the given (or a related) cancer, which can be input either manually (by a user) or automatically ex ⁇ tracted from other sources (e.g. reports).
  • additional data AI such as data re ⁇ ferring to the person's history (such as previous cases of ovarian and/or breast cancer of the person), the person's risk level (based on family history, in particular referring to previous cases of breast cancer in the person's family, ), the age of onset of the given (or a related) cancer, which can be input either manually (by a user) or automatically ex ⁇ tracted from other sources (e.g. reports).
  • a fifth step U comprises the processing U of the phenotypical features PF, PFi, PF 2 , PF 3 of the input image data ID through a classification and/or comparison algorithm based on the number of phenotypical reference features PRF and/or on the datasets DS .
  • the additional data AI can also be used as additional features which can be compared with ac- cordingly corresponding reference features from the same data collection that was used to provide the phenotypical refer ⁇ ence features.
  • a probability indicator value PI is derived from that processing U.
  • Random Forests classifier algorithm As example of underlying classification technology a trained Random Forests classifier algorithm is used. That Random For- est classifier has been trained with the phenotypical refer ⁇ ence features PRF so that the phenotypical features PF, PFi, PF 2 , PF 3 derived from the input image data ID can be proc ⁇ essed through the Random Forest classifier to receive the probability indicator value PI in the form of a confidence factor, i.e. a percentage value representing a probability of whether the person whose input image data ID have been used is a bearer of BRCA1 or BRCA2, i.e. that the given breast tu ⁇ mor 3 is linked to a genetic predisposition.
  • a confidence factor i.e. a percentage value representing a probability of whether the person whose input image data ID have been used is a bearer of BRCA1 or BRCA2, i.e. that the given breast tu ⁇ mor 3 is linked to a genetic predisposition.
  • the likelihood of a genetic vs. a sporadic breast cancer is assessed by means of a learning-based system, in this case a Random Forest classification algorithm.
  • the output is then produced, for instance in the form of a number (e.g. a per ⁇ centage or score) representing the confidence that the input sample in the form of the input image data ID corresponds to a genetically predisposed breast tumor 3 (as opposed to a sporadic tumor) .
  • high risk pa ⁇ tients with breast tumors 3 for which the method outputs an elevated confidence in genetic predisposition could be promptly referred to genetic counseling for further genetic tests and specific, targeted cancer management.
  • a clinician may choose to refer the patient to genetic counseling promptly.
  • a steep cost reduction can be achieved through avoidance of duplicate or unnecessary procedures (e.g. women who undergo breast-preserving surgery and, upon discovering high risk of genetic mutation, undergo double mastectomy) .
  • Fig. 6 shows a schematic block view of a tomographic imaging device 5 according to an embodiment of the invention. This will be described with reference to the example given in the preceding figures.
  • the tomographic imaging device 5 comprises an acquisition unit 7 for acquisition of the input image data ID and a de ⁇ termination assembly 9 according to an embodiment of the in ⁇ vention .
  • the determination assembly 9 comprises a first provision unit 11 in the form of a first input interface 11 which provides the input image data ID from the acquisition unit 7, i.e. from the tomographic imaging device 5. It further comprises a second provision unit 17, realized as a second input inter ⁇ face 17, which provides the phenotypical reference features PRF or the datasets DS from a database DB, which may be part of the tomographic imaging device 5 or connected to it via the second input interface 17.
  • the determination as- sembly 9 comprises an analysis unit 13 which extracts the phenotypical features PF, PFi, PF 2 , PF 3 from the input image data ID, a processing unit 15 which processes the phenotypi ⁇ cal features PF, PFi, PF 2 , PF 3 through the Random Forest clas- sification algorithm as described above and a derivement unit 19 which therefrom derives the probability indicator value PI. Via an output interface 21 this probability indicator value PI can then be transferred from the determination as- sembly 9 and from the tomographic imaging device 5 to a user, for instance to a computer monitor 23 where the user can receive the probability indicator value PI and therefrom decide further measures.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Theoretical Computer Science (AREA)
  • Medical Informatics (AREA)
  • General Health & Medical Sciences (AREA)
  • Biotechnology (AREA)
  • Chemical & Material Sciences (AREA)
  • Proteomics, Peptides & Aminoacids (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Genetics & Genomics (AREA)
  • Evolutionary Biology (AREA)
  • Biophysics (AREA)
  • Analytical Chemistry (AREA)
  • Spectroscopy & Molecular Physics (AREA)
  • Molecular Biology (AREA)
  • Nuclear Medicine, Radiotherapy & Molecular Imaging (AREA)
  • Radiology & Medical Imaging (AREA)
  • Quality & Reliability (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • General Physics & Mathematics (AREA)
  • Measuring And Recording Apparatus For Diagnosis (AREA)
  • Image Analysis (AREA)

Abstract

La présente invention concerne un procédé (Z) de détermination d'une valeur d'indicateur de probabilité (PI) représentant une probabilité d'une présence d'une prédisposition génétique d'un être pour une maladie spécifique. Le procédé (Z) comprend les étapes de fourniture (Y) d'un nombre de données d'image d'entrée (ID) d'un organe ou d'une structure qui présente potentiellement des effets de la maladie spécifique à partir d'un dispositif (5) d'imagerie tomographique, d'analyse (X) des données d'image d'entrée (ID) par extraction à partir de celles-ci d'un nombre de caractéristiques phénotypiques de niveau faible (PF, PF1, PF2, PF3), de fourniture (W) d'une pluralité de caractéristiques de référence phénotypiques (PRF) issues de données d'image de référence du même organe ou de la même structure d'un nombre d'êtres de référence et/ou d'une base de données (DB) ayant un nombre d'ensembles de données (DS) représentant les caractéristiques de référence phénotypiques (PRF), la pluralité de caractéristiques de référence phénotypiques (PRF) comprenant de telles caractéristiques de référence phénotypiques (PRF) qui correspondent aux caractéristiques phénotypiques (PF, PF1, PF2, PF3) des données d'image d'entrée (ID), de traitement (U) des caractéristiques phénotypiques (PF, PF1, PF2, PF3) des données d'image d'entrée (ID) par l'intermédiaire d'un algorithme de classification et/ou de comparaison sur la base du nombre de caractéristiques de référence phénotypiques (PRF) et/ou sur la base des ensembles de données (DS), de déduction (T) de la valeur d'indicateur de probabilité (PI) à partir du traitement (U). La présente invention concerne également un ensemble de détermination (9) à cette fin.
PCT/EP2013/077309 2012-12-20 2013-12-19 Détermination d'une valeur d'indicateur de probabilité WO2014096147A2 (fr)

Priority Applications (1)

Application Number Priority Date Filing Date Title
EP13819022.8A EP2901340A2 (fr) 2012-12-20 2013-12-19 Détermination d'une valeur d'indicateur de probabilité

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
EP12198642.6 2012-12-20
EP12198642 2012-12-20

Publications (2)

Publication Number Publication Date
WO2014096147A2 true WO2014096147A2 (fr) 2014-06-26
WO2014096147A3 WO2014096147A3 (fr) 2014-10-02

Family

ID=49949634

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/EP2013/077309 WO2014096147A2 (fr) 2012-12-20 2013-12-19 Détermination d'une valeur d'indicateur de probabilité

Country Status (2)

Country Link
EP (1) EP2901340A2 (fr)
WO (1) WO2014096147A2 (fr)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
DE102014213828A1 (de) * 2014-07-16 2016-01-21 Siemens Aktiengesellschaft Prognosesystem und Prognoseverfahren

Family Cites Families (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6282305B1 (en) * 1998-06-05 2001-08-28 Arch Development Corporation Method and system for the computerized assessment of breast cancer risk
EP2923336B1 (fr) * 2012-11-20 2017-12-13 Koninklijke Philips N.V. Phénotypage intégré utilisant des caractéristiques de texture d'image

Non-Patent Citations (16)

* Cited by examiner, † Cited by third party
Title
ANTONIOU, A. ET AL.: "Average Risks of Breast and Ovarian Cancer Associated with BRCA1 or BRCA2 Mutations Detected in Case Series Unselected for Family History: A Combined Analysis of 22 Studies", AM J HUM GENET, vol. 72, no. 5, 2003, pages 1117 - 1130
BREIMAN L.; RANDOM FORESTS, MACHINE LEARNING, vol. 45, no. 1, 2001, pages 5 - 32
DALY, MARY B. ET AL.: "Genetic/Familial High-Risk Assessment: Breast and Ovarian", JOURNAL OF THE NATIONAL COMPREHENSIVE CANCER NETWORK., vol. 8, no. 5, 2010, pages 562 - 594
DIAMANDIS, MARIA ET AL.: "Personalized Medicine: Marking a New Epoch in Cancer Patient Management", MOL CANCER RES, vol. 8, no. 9, 2010, pages 1175 - 1187
EVANS, D. G. R. ET AL.: "A new scoring system for the chances of identifying a BRCA1/2 mutation outperforms existing models including BRCAPRO", J MED GENET, vol. 41, no. 6, 2004, pages 474 - 80
FRANK THOMAS S. ET AL.: "Clinical Characteristics of Individuals With Germline Mutations in BRCA1 and BRCA2: Analysis of 10,000 Individuals", JOURNAL OF CLINICAL ONCOLOGY, vol. 20, no. 6, 2002, pages 1480 - 1490
HU, MING-KUEI: "Visual Pattern Recognition by Moment Invariants", IRE TRANSACTIONS ON INFORMATION THEORY, 1962, pages 179 - 187, XP011217262, DOI: doi:10.1109/TIT.1962.1057692
NATURE, vol. 434, 14 April 2005 (2005-04-14), pages 917 - 921
PARMIGIANI, GIOVANNI ET AL.: "Determining Carrier Probabilities for Breast Cancer-Susceptibility Genes BRCA1 and BRCA2", AM J HUM GENET, vol. 62, 1998, pages 145 - 158
PEJNOVIC, P. ET AL.: "Object Recognition by Invariants", PROC INT CONF ON PATTERN RECOGNITION, 1992, pages 434 - 437, XP010029980, DOI: doi:10.1109/ICPR.1992.201811
PETRUCELLI, NANCIE ET AL.: "Hereditary breast and ovarian cancer due to mutations in BRCA1 and BRCA2", GENETICS IN MEDICINE, vol. 12, no. 5, 2010, pages 245 - 259
PIERCE, LORI: "Radiotherapy for Breast Cancer in BRCA1/BRCA2 Carriers: Clinical Issues and Management Dilemmas", SEMIN RADIAT ONCOL, vol. 12, no. 4, 2002, pages 352 - 361, XP005454681, DOI: doi:10.1053/srao.2002.35254
SABEL, MICHAEL S.: "Is Breast Conservation the Wrong Choice in Women With BRCA1 or BRCA2 Mutations?", AM J ONCOL REV, vol. 1, no. 3, 2002, pages 155 - 161
See also references of EP2901340A2
TYRER, JONATHAN ET AL.: "A breast cancer prediction model incorporating familial and personal risk factors", STATIST MED, vol. 23, no. 7, 2004, pages 1111 - 1130
VAN DER GROEP, P. ET AL.: "Distinction between hereditary and sporadic breast cancer on the basis of clinicopathological data", J CLIN PATHOL, vol. 59, 2006, pages 611 - 617

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
DE102014213828A1 (de) * 2014-07-16 2016-01-21 Siemens Aktiengesellschaft Prognosesystem und Prognoseverfahren

Also Published As

Publication number Publication date
EP2901340A2 (fr) 2015-08-05
WO2014096147A3 (fr) 2014-10-02

Similar Documents

Publication Publication Date Title
US10910094B2 (en) Computer-based diagnostic system
US11037070B2 (en) Diagnostic test planning using machine learning techniques
Eriksson et al. Identification of women at high risk of breast cancer who need supplemental screening
Wanders et al. Interval cancer detection using a neural network and breast density in women with negative screening mammograms
JP2017515574A (ja) 症例の難しさに基づくコンピュータ支援患者層別化のための方法及びシステム
KR20210095654A (ko) 영상을 사용하여 유방암 위험을 평가하기 위한 시스템 및 방법
JP2009516551A (ja) 医用画像の定量的および定性的なコンピュータ支援分析方法およびシステム
US8929625B2 (en) Method and device for side-effect prognosis and monitoring
EP3685401B1 (fr) Procédé et appareil de dérivation d'un ensemble de données d'apprentissage
CN115362470A (zh) 用于预测未来肺癌的风险的系统和方法
EP3573072A1 (fr) Réalisation d'une évaluation prognostique
US20180053297A1 (en) Methods and Apparatuses for Detection of Abnormalities in Low-Contrast Images
JP2017532998A (ja) 縦断的特徴に基づく関心組織の健康状態の分類
Aggarwal et al. Semantic and content-based medical image retrieval for lung cancer diagnosis with the inclusion of expert knowledge and proven pathology
AU2022375759A1 (en) Systems and methods to process electronic images for determining treatment
Macchia et al. Multidisciplinary tumor board smart virtual assistant in locally advanced cervical cancer: A proof of concept
Carney et al. Likelihood of additional work-up among women undergoing routine screening mammography: the impact of age, breast density, and hormone therapy use
WO2014096147A2 (fr) Détermination d'une valeur d'indicateur de probabilité
Nivetha et al. Lung cancer detection at early stage using PET/CT imaging technique
EP4002382A1 (fr) Utilisation de données médicales temporelles non structurées pour la prédiction de la maladie
Muis et al. CNN-based Approach for Enhancing Brain Tumor Image Classification Accuracy
Zhang et al. Value of rehabilitation training for children with cerebral palsy diagnosed and analyzed by computed tomography imaging information features under deep learning
Caramihai et al. Evaluation of breast cancer risk by using fuzzy logic
US20210343004A1 (en) Method for Digitally Generating Scores for Multiple Diagnostic Tests from Tissue Assayed with a Single Test
EP4071768A1 (fr) Dispositif cad et procédé d'analyse d'images médicales

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 13819022

Country of ref document: EP

Kind code of ref document: A2

WWE Wipo information: entry into national phase

Ref document number: 2013819022

Country of ref document: EP