Systems and methods for identifying molecular groups of meningioma using radiographic and radiomic features

Abstract

The present disclosure pertains to a method of predicting a molecular subtype of a tumor of a subject by (1) receiving preoperative imaging data of the tumor; (2) extracting one or more features of the imaging data; and (3) correlating the extracted features of the imaging data to a molecular subtype of the tumor. The methods may also include steps of outputting the molecular subtype of the tumor and implementing a treatment decision. The present disclosure also pertain to a computing device and a diagnostic test for predicting a molecular subtype of a tumor of a subject in accordance with the aforementioned methods.

Description

TITLESYSTEMS AND METHODS FOR IDENTIFYING MOLECULAR GROUPS OF MENINGIOMA USING RADIOGRAPHIC AND RADIOMIC FEATURESCROSS-REFERENCE TO RELATED APPLICATIONS

[0001] This application claims priority to U.S. Provisional Patent Application No. 63 / 661,258, filed on June 18, 2024. The entirety of the aforementioned application is incorporated herein by reference.BACKGROUND

[0002] Current methods and systems for differentiating between different types of tumors (e.g., meningiomas) suffer from numerous limitations. Numerous embodiments of the present disclosure aim to address the aforementioned limitations.SUMMARY

[0003] In some embodiments, the present disclosure pertains to a method of predicting a molecular subtype of a tumor of a subject. In some embodiments, the methods of the present disclosure include: (1) receiving preoperative imaging data of the tumor; (2) extracting one or more features of the imaging data; and (3) correlating the extracted features of the imaging data to a molecular subtype of the tumor. In some embodiments, the methods of the present disclosure also include a step of outputting the molecular subtype of the tumor. In some embodiments, the methods of the present disclosure also include a step of implementing a treatment decision. In some embodiments, the treatment decision includes monitoring the course of the tumor, implementing a tumor treatment regimen, implementing a tumor management regimen , or combinations thereof.

[0004] Additional embodiments of the present disclosure pertain to a computing device for predicting a molecular subtype of a tumor of a subject. In some embodiments, the computing device includes one or more computer readable storage mediums having a program code embodied therewith. In some embodiments, the program code includes: (1) programming instructions for receiving preoperative imaging data of the tumor; (2) programming instructions for extracting one or more features of the imaging data; and (3) programming instructions for correlating the extracted features of the imaging data to a molecular subtype of the tumor, hi some embodiments, the program code also includes programming instructions for outputting the molecular subtype of the tumor. In some embodiments, the program code also includes programming instructions for recommending a treatment decision.

[0005] In some embodiments, the computing device is a component of a diagnostic test for predicting a molecular subtype of a tumor of a subject. Additional embodiments of the present disclosure pertain to a diagnostic test for use in predicting a molecular subtype of a tumor of a subject. In some embodiments, the diagnostic test includes a computing device of the present disclosure.DRAWINGS

[0006] FIG. 1A illustrates a method of predicting a molecular subtype of a tumor of a subject in accordance with various embodiments of the present disclosure.

[0007] FIG. IB illustrates a computing device for predicting a molecular subtype of a tumor of a subject.

[0008] FIGS. 2A-2D illustrate that pre MRI information is insufficient to predict outcome. Shown are receiver operating characteristics (ROC) using models generated from variables collected before imaging. These characteristics were unable to differentiate between benign and aggressive meningiomas with an area under the receiver operating characteristic curve (AUROC) of 0.57-0.59 for a neural network or a random forest algorithm.

[0009] FIGS. 3A-3J illustrate examples of ten chosen variables used in a study in Example 1. The quality of each variable associated with MenG A / B (non-aggressive) vs. MenG C (aggressive) tumors is stated along with a representative image. The predictive power of each variable in isolation on the training set (n=178) is depicted to the right showing that none of the variables alone is sufficient. Only the weight of the combined ten values can predict MenG C / aggressiveness preoperatively.

[0010] FIGS. 4A-4C show graphs of model predictions. ROC was graphed using the training set tumors (n=178) (FIG. 4A), a second, unknown validation set (n=66) (FIG. 4B), and all tumors (n=244) (FIG. 4C). For each graph, there are two models derived and / or evaluated from a singlelayered neural network and a random forest algorithm. Maximizing performance using 10-repeat 10- fold internal cross-validation resulted in an AUROC of 0.84 and 0.85 (FIG. 4A), 0.88 and 0.89 (FIG. 4B) and both 0.86 (FIG. 4C), respectively.

[0011] FIG. 5 illustrates feature selections of patients from medical records. A total of 133 features were extracted from medical records and MRI sequences. Of these, 75 trended to be positively correlative with MenG C / aggressive status, and 10 were ultimately selected after recursive feature elimination using a rudimentary random forest algorithm to maximize the AUROC.

[0012] FIGS. 6A-6D illustrate potential courses of treatment for patients based on MenG prediction.DETAILED DESCRIPTION

[0013] It is to be understood that both the foregoing general description and the following detailed description are illustrative and explanatory, and are not restrictive of the subject matter, as claimed. In this application, the use of the singular includes the plural, the word “a” or “an” means “at least one”, and the use of “or” means “and / or”, unless specifically stated otherwise. Furthermore, the use of the term “including”, as well as other forms, such as “includes” and “included”, is not limiting. Also, terms such as “element” or “component” encompass both elements or components comprising one unit and elements or components that include more than one unit unless specifically stated otherwise.

[0014] The section headings used herein are for organizational purposes and are not to be construed as limiting the subject matter described. All documents, or portions of documents, cited in this application, including, but not limited to, patents, patent applications, articles, books, and treatises, are hereby expressly incorporated herein by reference in their entirety for any purpose. In the event that one or more of the incorporated literature and similar materials defines a term in a manner that contradicts the definition of that term in this application, this application controls.

[0015] Current methods and systems for differentiating between different types of tumors suffer from numerous limitations. For instance, meningiomas are the most common primary intracranial neoplasms. Current methods for grading meningiomas rely on a grading system defined by the World Health Organization (WHO). This grading scheme is based predominantly on histologic features and now limited molecular characterization, which docs not consistently reflect the biologic behavior of meningiomas. This leads to both under-treatment and over-treatment of patients, and hence, suboptimal outcomes.

[0016] Studies have shown that classifications based on transcriptomic, cytogenetic, and methylomic signatures better characterize and prognosticate meningioma activity as compared with traditional histopathalogical grading. Using current technology, such classifications cannot be made preoperatively.

[0017] Accordingly, there is a need in the art for systems and methods that improve existing methodologies by providing preoperative prognosis of tumors, such as meningiomas. Numerous embodiments of the present disclosure aim to address this need.

[0018] In some embodiments, the present disclosure pertains to a method of predicting a molecular subtype of a tumor of a subject. In some embodiments illustrated in FIG. 1A, the methods of the present disclosure include: receiving preoperative imaging data of the tumor (step 10); extracting one or more features of the imaging data (step 11); and correlating the extracted features of the imaging data to a molecular subtype of the tumor (step 12). In some embodiments, the methods of the present disclosure also include a step of outputting the molecular subtype of the tumor (step 13). In some embodiments, the methods of the present disclosure also include a step of implementing a treatment decision (step 14). In some embodiments, the treatment decision includes monitoring the course of the tumor (step 15), implementing a tumor treatment regimen (step 16), implementing a tumor management regimen (step 17), or combinations thereof.

[0019] Additional embodiments of the present disclosure pertain to a computing device for predicting a molecular subtype of a tumor of a subject. In some embodiments, the computing device includes one or more computer readable storage mediums having a program code embodied therewith. In some embodiments, the program code includes: (1) programming instructions for receiving preoperative imaging data of the tumor; (2) programming instructions for extracting one or more features of the imaging data; and (3) programming instructions for correlating the extracted features of the imaging data to a molecular subtype of the tumor, hr some embodiments, the program code also includes programming instructions for outputting the molecular subtype of the tumor. In some embodiments, the program code also includes programming instructions for recommending a treatment decision.

[0020] In some embodiments, the computing device is a component of a diagnostic test for predicting a molecular subtype of a tumor of a subject. Additional embodiments of the present disclosure pertain to a diagnostic test for use in predicting a molecular subtype of a tumor of a subject. In some embodiments, the diagnostic test includes a computing device of the present disclosure.

[0021] As set forth in more detail herein, the methods, computing devices and diagnostic tests of the present disclosure can have numerous embodiments.

[0022] Tumors

[0023] The methods, computing devices and diagnostic tests of the present disclosure may be utilized to predict a molecular subtype of various tumors. For instance, in some embodiments, the tumor includes a brain tumor. In some embodiments, the tumor includes meningioma. In some embodiments, the tumor includes schwannoma.

[0024] Subjects

[0025] The methods, computing devices and diagnostic tests of the present disclosure may be utilized to predict a molecular subtype of tumors of various subjects. For instance, in some embodiments, the subject is a human being. In some embodiments, the subject is a non-human mammal, such as a dog or a cat.

[0026] Preoperative imaging data

[0027] The methods, computing devices and diagnostic tests of the present disclosure may receive various preoperative imaging data. For instance, in some embodiments, the preoperative imaging data includes radiological imaging data, computed tomography (CT) scan data, positron emission tomography (PET) data, magnetic resonance imaging (MRI) data, age of the subject, gender of the subject, location of the subject, or combinations thereof. In some embodiments, the preoperative imaging data includes magnetic resonance imaging (MRI) data (e.g., data from MRI with or without contrast). In some embodiments, the preoperative imaging data may include an MRI image of a human brain having a brain tumor.

[0028] Extraction of features from imaging data

[0029] The methods, computing devices and diagnostic tests of the present disclosure may extract various features from preoperative imaging data. For instance, in some embodiments, such data may include different sequences. In some embodiments, the extracted features of the imaging data include, without limitation, radiomic features, radiographic features, first order features, quantitative features, qualitative features, information related to tumor location, information related to tumor length, subject’s gender, information related to tumor volume, information related to tumor edema, supra- vs. infratentorial status of the tumor, presence of multiple tumors, local bony changes, enhancement pattern of the tumor, heterogenous enhancement pattern of the tumor, cystic degeneration of the tumor, central necrosis of the tumor, presence of intra-tumoral flow voids, tumor intensity on T1 / T2 sequences, tumor restriction on apparent diffusion coefficient (ADC) sequence, tumor shape, distinct tumor margins on T2 (“CSF cleft”), tumor hypointensity on T1 non-contrast sequences, indistincttumor margins, lmc2 (i.e., a radiomics feature that provides insights of the texture of a tumor), dependence entropy (i.e., a specific type of a radiomic feature that is derived from the Gray Level Dependence Matrix (GLDM) and measures the randomness or uncertainty within gray level dependencies, where a higher dependence entropy suggests a more heterogeneous texture pattern in an image), correlation (i.e., a feature that refers to the statistical relationship between different radiomic features or between radiomic features and other variables, such as clinical parameters, genomic data, or treatment response), appearance of dural tail on the tumor, susceptibility changes of the tumor on gradient or fast field echo sequences, or combinations thereof.

[0030] In some embodiments, the extracted features include, without limitation, tumor location, information related to tumor length, subject’s gender, supra- vs. infratentorial status of the tumor, tumor shape, tumor hypointensity on T1 non-contrast sequences, indistinct tumor margins, lmc2, dependence entropy, correlation, or combinations thereof. In some embodiments, the extracted features include radiomic features of the imaging data. In some embodiments, the extracted features include radiographic features of the imaging data. In some embodiments, the extracted features include first order features of the imaging data.

[0031] In some embodiments, the extracted features include segmented features of the imaging data. In some embodiments, the extracted features include information related to tumor edema. In some embodiments, information related to tumor edema may be categorized as mild (crescent), moderate (lobar with or without local mass effect), or severe (midline shift).

[0032] In some embodiments, the extracted features include, without limitation, supratentorial status, parafalcine location, larger volume, edema, heterogenous enhancement, hypointensity on T1 noncontrast sequence, and indistinct tumor margins. In some embodiments, the extracted features may include information related to tumor location. In some embodiments, information related to tumor location may be grouped into categories that include convexity, skull base (e.g., olfactory groove, planum sphenoidale / tuberculum sellae, sellar / suprasellar, clinoidal, sphenoid wing, and / or cavernous sinus), parafalcine (e.g. , falcine and / or parasagittal), posterior fossa (e.g. , petroclival, cerebellopontine angle, clival / foramen magnum, tentorial, and / or cerebellar), and / or intraventricular.

[0033] In some embodiments, the extracted features may include information related to tumor volume. In some embodiments, tumor volume may be calculated using the maximum dimensions of the tumor.

[0034] In some embodiments, the extracted features include local bony changes. In some embodiments, the local bony changes include hyperostosis and invasion. In some embodiments, the local bony changes may suggest calcification as opposed to hemorrhage.

[0035] In some embodiments, the extracted features include information related to tumor length. In some embodiments, information related to tumor length includes a tumor’s major axis length.

[0036] In some embodiments, the extracted features include tumor shape. In some embodiments, the tumor shape includes flatness.

[0037] The methods, computing devices and diagnostic tests of the present disclosure may extract features from preoperative imaging data in various manners. For instance, in some embodiments, the extraction occurs in a manual manner. In some embodiments, the data may be processed by manual annotation to extract features.

[0038] In some embodiments, the extraction occurs in a semi-automatic manner. In some embodiments, the extraction includes semi-automatic segmentation of the imaging data.

[0039] In some embodiments, tumors may be semi-automatically segmented from preoperative T1 postcontrast MRIs. In some embodiments, such images may be normalized and resampled for homogenous comparison across scans of different slice thicknesses and acquisition patterns. In some embodiments, shape and first order features may be extracted.

[0040] In some embodiments, summary statistics may be calculated for the extracted features. Variables may be weighted and categorized based on their clinical significance for prognostication of molecular subtypes of tumors. Statistical methods applied may include, without limitation, Pearson's Chi-squared, and Wilcoxon rank sum test and normalization. In some embodiments, the extracted features may be used to train machine learning models.

[0041] Correlation to a molecular' subtype of the tumor

[0042] The methods, computing devices and diagnostic tests of the present disclosure may correlate the extracted features of an imaging data to various molecular subtypes of a tumor. For instance, in some embodiments, the correlation includes correlation to a benign subtype of a tumor. In some embodiments, the correlation includes correlation to a malignant subtype of a tumor.

[0043] In some embodiments, the correlation includes correlation to a benign or malignant molecular subtype of meningioma. In some embodiments, the correlation includes classification of extractedimaging data into a molecular subtype of meningioma group A (MenG A), meningioma group B (MenG B), meningioma group C (MenG C), or combinations thereof.

[0044] In some embodiments, the correlation occurs through the utilization of a machine-learning algorithm trained on the extracted features of the image data. In some embodiments, the machine learning algorithm includes a neural network model, a random forest model, a machine learning based classifier, a decision tree classifier, a boosted decision tree classifier, a single-layered neural network, or combinations thereof. In some embodiments, the correlation includes classification of extracted and segmented features of the imaging data into at least two groups using at least one machine learning based classifier.

[0045] In some embodiments, machine learning algorithms may be initially trained on preoperative imaging data from subjects where resected tumor data may be available. In further embodiments, neural network models may be used to classify tumors into molecular subtypes based on extracted features of the imaging data (e.g., radiographic and radiomic features extracted from routine preoperative imaging).

[0046] In some embodiments, tumors may be semi-automatically segmented from preoperative T1 postcontrast MRIs. Images may be normalized and resampled for homogenous comparison across scans of different slice thicknesses and acquisition patterns.

[0047] Treatment decision

[0048] In some embodiments, the methods of the present disclosure also include a step of implementing a treatment decision. In some embodiments, the computing devices of the present disclosure also include programming instructions for recommending a treatment decision.

[0049] In some embodiments, the treatment decision includes monitoring the course of the tumor, implementing a tumor treatment regimen, implementing a tumor management regimen, or combinations thereof.

[0050] In some embodiments, the treatment decision includes implementing a tumor treatment regimen, removing the tumor, or combinations thereof. In some embodiments, the tumor treatment regimen includes administering a therapeutic agent to the subject. In some embodiments, the tumor treatment regimen includes removing the tumor from the subject.

[0051] Applications and variations

[0052] The methods, computing devices and diagnostic tests of the present disclosure may have various applications. For instance, in some embodiments, the methods of the present disclosure occur in a computer-implemented manner. In some embodiments, the methods, computing devices and diagnostic tests of the present disclosure may be used to prognosticate the aggressiveness of a tumor. In some embodiments, the methods of the present disclosure include a computer implemented method for prognosticating the aggressiveness of a tumor through the following steps: (a) receiving on at least one processor, at least one preoperative image from a subject with a brain tumor; (b) performing semiautomatic segmentation of the preoperative image; (c) classifying the segmented image into at least two groups using at least one machine learning based classifier; and (d) outputting, using at least one processor, a classification of the sample concerning prognosis of the tumor in the subject based on steps (a)-(c). In some embodiments, the preoperative image is a magnetic resonance image (MRI). In some embodiments, the semi-automatic segmentation includes extraction of first order features. In some embodiments, the extracted data includes radiomic features, radiographic features, or combinations thereof. In some embodiments, the tumor is a meningioma in a human brain. In some embodiments, the classification includes a molecular subtype classification as a MenG A, MenG B, MenG C, or combinations thereof. In some embodiments, the classifier includes a boosted decision tree model, a neural network model, or combinations thereof.

[0053] The methods, computing devices and diagnostic tests of the present disclosure may operate under various platforms. For instance, in some embodiments, the computing devices and diagnostic tests of the present disclosure may be in the form of an app or a software where a subject (e.g., patient) can upload their imaging data for evaluation. In some embodiments, the computing devices and diagnostic tests of the present disclosure may be integrated into an existing software or system, such as an Epic or a PACS system that allows radiologists to predict a molecular subtype of a tumor.

[0054] Computing devices

[0055] The computing devices of the present disclosure can include various types of computer- readable storage mediums. For instance, in some embodiments, the computer-readable storage medium can be a tangible device that can retain and store instructions for use by an instruction execution device. In some embodiments, the computer-readable storage medium may include, without limitation, an electronic storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or combinations thereof.

[0056] A non-exhaustive list of more specific examples of suitable computer-readable storage mediums includes, without limitation, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), a static random access memory (SRAM), a portable compact disc read-only memory (CD-ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanically encoded device, or combinations thereof.

[0057] A computer-readable storage medium, as used herein, is not to be construed as being transitory signals per se. Such transitory signals may be represented by radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through a waveguide or other transmission media (e.g., light pulses passing through a fiber-optic cable), or electrical signals transmitted through a wire.

[0058] In some embodiments, computer-readable program instructions for computing devices can be downloaded to respective computing / processing devices from a computer- readable storage medium or to an external computer or external storage device via a network, such as the Internet, a local area network (LAN), a wide area network (WAN) and / or a wireless network. In some embodiments, the network may include copper transmission cables, optical transmission fibers, wireless transmission, routers, firewalls, switches, gateway computers and / or edge servers. In some embodiments, a network adapter card or network interface in each computing / processing device receives computer-readable program instructions from the network and forwards the computer-readable program instructions for storage in a computer-readable storage medium within the respective computing / processing device.

[0059] In some embodiments, computer-readable program instructions for carrying out operations of the present disclosure may be assembler instructions, instruction- sct-architccturc (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state-setting data, configuration data for integrated circuitry, or either source code or object code written in any combination of one or more programming languages, including an object-oriented programming language such as Smalltalk, C++, or the like, and procedural programming languages, such as the "C" programming language or similar programming languages.

[0060] In some embodiments, the computer-readable program instructions may execute entirely on the user's computer, partly on the user's computer, as a stand-alone software package, partly on the user's computer and partly on a remote computer or entirely on the remote computer or server. In thelatter scenario, the remote computer may be connected in some embodiments to the user's computer through any type of network, including a LAN or a WAN, or the connection may be made to an external computer (for example, through the Internet using an Internet Service Provider). In some embodiments, electronic circuitry including, for example, programmable logic circuitry, field- programmable gate arrays (FPGA), or programmable logic arrays (PL A) may execute the computer- readable program instructions by utilizing state information of the computer-readable program instructions to personalize the electronic circuitry in order to perform aspects of the present disclosure.

[0061] Embodiments of the present disclosure for predicting a molecular subtype of a tumor of a subject as discussed herein may be implemented using a computing device illustrated in FIG. IB. Referring now to FIG. IB, FIG. IB illustrates an embodiment of the present disclosure of the hardware configuration of a computing device 30 which is representative of a hardware environment for practicing various embodiments of the present disclosure.

[0062] Computing device 30 has a processor 31 connected to various other components by computing device bus 32. An operating system 33 runs on processor 31 and provides control and coordinates the functions of the various components of FIG. IB. An application 34 in accordance with the principles of the present disclosure runs in conjunction with operating system 33 and provides calls to operating system 33, where the calls implement the various functions or services to be performed by application 34. Application 34 may include, for example, a program for predicting a molecular subtype of a tumor of a subject as discussed in the present disclosure.

[0063] Referring again to FIG. IB, read-only memory ("ROM") 35 is connected to computing device bus 32 and includes a basic input / output computing device ("BIOS") that controls certain basic functions of computing device 30. Random access memory ("RAM") 36 and disk adapter 37 arc also connected to computing device bus 32. It should be noted that software components including operating system 33 and application 34 may be loaded into RAM 36, which may be computing device’s 30 main memory for execution. Disk adapter 37 may be an integrated drive electronics ("IDE") adapter that communicates with a disk unit 38 (e.g., a disk drive). It is noted that the program for predicting a molecular subtype of a tumor of a subject, as discussed in the present disclosure.

[0064] Computing device 30 may further include a communications adapter 39 connected to computing device bus 32. Communications adapter 39 interconnects computing device bus 32 with an outside network (e.g., wide area network) to communicate with other devices.

[0065] Aspects of the present invention are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and systems according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams and combinations of blocks in the flowchart illustrations and / or block diagrams can be implemented by computer-readable program instructions.

[0066] These computer-readable program instructions may be provided to a processor of a computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions / acts specified in the flowchart and / or block diagram block or blocks. These computer-readable program instructions may also be stored in a computer- readable storage medium that can direct a computer, a programmable data processing apparatus, and / or other devices to function in a particular manner, such that the computer-readable storage medium having instructions stored therein includes an article of manufacture including instructions which implement aspects of the function / act specified in the flowchart and / or block diagram block or blocks.

[0067] The computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other device to produce a computer- implemented process, such that the instructions which execute on the computer, other programmable apparatus, or other device implement the functions / acts specified in the flowchart and / or block diagram block or blocks.

[0068] The flowchart and block diagrams in the Figures illustrate the architecture, functionality, and operation of possible implementations of computing devices, methods, and computing devices according to various embodiments of the present disclosure. In this regard, each block in the flowchart orblock diagrams may represent a module, segment, or portion of instructions, which includes one or more executable instructions for implementing the specified logical function(s). In some alternative implementations, the functions noted in the blocks may occur out of the order noted in the Figures. For example, two blocks shown in succession may, in fact, be accomplished as one step, executed concurrently, substantially concurrently, in a partially or wholly temporally overlapping manner, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and / or flowchart illustration, andcombinations of blocks in the block diagrams and / or flowchart illustration, can be implemented by special purpose hardware-based computing devices that perform the specified functions or acts or carry out combinations of special purpose hardware and computer instructions.

[0069] Applications and advantages

[0070] The methods, computing devices and diagnostic tests of the present disclosure can provide numerous advantages. For instance, in some embodiments, the methods, computing devices and diagnostic tests of the present disclosure provide a facile and convenient process for predicting a molecular subtype of a tumor from preoperative imaging data, which can in turn improve surgical planning and postoperative care for patients.

[0071] As such, the methods, computing devices and diagnostic tests of the present disclosure can provide numerous advantages. For instance, in some embodiments, the methods, computing devices and diagnostic tests of the present disclosure can be utilized in healthcare facilities with limited resources to accurately predict molecular subtypes of tumors. Such healthcare facilities can include smaller hospitals or healthcare centers in under-developed regions of the world where sequencing facilities may be scarce but MRIs are feasible. In some embodiments, the methods, computing devices and diagnostic tests of the present disclosure can be utilized as cost-effective tools to identify MenG C tumors.

[0072] Additional embodiments

[0073] Reference will now be made to more specific embodiments of the present disclosure and experimental results that provide support for such embodiments. However, Applicant notes that the disclosure below is for illustrative purposes only and is not intended to limit the scope of the claimed subject matter in any way.

[0074] Example 1. Pre-operative MRI-based radiomics predicts aggressive meningiomas to guide pre-operative planning and surgical care

[0075] Meningiomas fall into three distinct biological classes (A, B, C), with the first two being benign and the third invariably aggressive. This molecular classification system outperforms histopathological (WHO) grading in predicting tumor behavior but is not yet part of routine practice. Moreover, the classification requires a tumor sample, and not all hospitals have the capacity for molecular analysis.

[0076] All meningioma patients, however, will undergo MRI as part of their diagnostic workup. Accordingly, Applicant sought radiographic and radiomic features that predict type C tumors to improve management. Applicant’s neural network model had an 84.5% accuracy with an area under the curve (AUC) of 0.86, a perturbation projection vector (PPV) of 71%, and a neural path value (NPV) of 90%. Applicant’s random forest model had an 81.6% accuracy with AUC of 0.81, PPV of 64.5%, and NPV of 89%. This noninvasive MRI-based approach can rule out benign meningioma and pinpoint patients who require more aggressive treatment as well as stratify participants for clinical trials.

[0077] For the experimental design, Applicant molecularly categorized tumor samples resected from patients (n=178, training set) as either Meningioma Groups (MenG) A or B (biologically distinct but both benign) or MenG C (invariably aggressive). Applicant then analyzed the patients’ preoperative MRIs, segmented them on T1 post-contrast and T2 FLAIR sequences, and extracted quantitative radiomic features that Applicant then used to train neural network and random forest algorithms. These algorithms were tested on another group of patients (n=66) as a validation set.

[0078] The results indicated that both classification algorithms predicted aggressive molecular status with similar threshold accuracy (Training set: 79-80%, AUC 0.84-0.85; Validation set: 83-89%, AUC 0.88-0.89). The overall predictive ability of the models on all tumors (n=244) was equally reliable (accuracy 82%, AUC 0.86). Importantly, the random forest algorithm made no false negative errors when evaluating meningiomas without edema. Within the validation cohort, radiomics would have altered pre-surgical decision-making in 38% of patients. As such, high-risk meningiomas can be predicted by routine preoperative MRI, leading to better preoperative counseling and treatment planning.

[0079] Example 1.1. Background

[0080] Meningiomas account for 41% of all primary intracranial neoplasms. Since the majority are indolent-in fact, some are detected incidentally during scanning for some unrelated condition-the current standard of care is to follow the patient with serial imaging unless the tumor is causing symptoms, threatening to compress critical structures, or demonstrating growth, in which case it will be surgically resected. Unfortunately, 20-40% of meningiomas recur after surgery, including up to one-fifth of the tumors that the World Health Organization (WHO) histopathological grading criteria would classify as benign.

[0081] There have thus been several attempts to extract magnetic resonance imaging (MRI) features that might predict the likelihood of invasive behavior, based on the success of this approach with other brain tumor types. Aggressive meningiomas have been associated to varying degrees with tumor volume, tumor necrosis, irregular margins, and peritumoral edema. However, none of these features reliably predicted tumor grade. Moreover, some of the studies lacked the necessary sample size to generalize their findings by having a great imbalance between groups. Other studies relied entirely on internal cross-validation. Moreover, some focused on subsets of meningiomas such as only highgrade meningioma. Additionally, some studies included subtotal resections in the definition of recurrence, which is likely a re-growth rather than a true recurrence. The greatest challenge for these studies, however, was their focus on preoperatively predicting WHO histopathological grade, which, as noted above, does not reliably predict aggressive behavior.

[0082] The need to better identify invasive meningioma led Applicant to apply molecular profiling techniques to develop a more accurate biological classification system. Using tumors resected from a sizable patient cohort in the Texas Medical Center, Applicant discovered that meningiomas can be divided into three biological classes distinguished by their RNA-seq signatures, which Applicant called Meningioma Groups (MenG) A, B, and C. Both MenG A and B tumors are remarkably indolent and differ primarily by their NF2 status, whereas MenG C tumors recur repeatedly despite total resection and regardless of their WHO grade (1, 2, or 3).

[0083] There are also some hints as to biological class from tumor location and certain demographic features such as sex. Applicant subsequently showed that these groups can be identified by methylomic, transcriptomic, or cytogenetic profiling and are conserved across species. Molecular classification therefore not only predicts recurrence risk with far greater accuracy than histopathology, but it also explains apparent inconsistencies in previous studies that relied on WHO grading.

[0084] The aforementioned results prompted Applicant to revisit the possibility that radiomics might be able to distinguish aggressive tumors on preoperative MRI-based on accurate molecular classification rather than WHO grade.

[0085] Methods to preoperatively predict disease course are gaining in parallel with advances in data and imaging analytics, and radiogenomics has proven capable of linking imaging phenotypes with genomic signatures for other malignant CNS tumors, including glioma and medulloblastoma. Here,Applicant aims to predict this molecular classification system preoperatively using quantitative radiomic and clinico-radiographic features.

[0086] Example 1,2, Tumor Selection

[0087] Intracranial tumor samples resected between 2012 and 2022 at Baylor College of Medicine with both a high-quality preoperative MRI and that had undergone MenG classification were included. The model training cohort consisted of 178 tumors (2012-2020) and the validation cohort of 66 tumors (FIGS. 2A-2D).

[0088] Example 1,3. Pre-MRI Data Retrieval

[0089] For each included tumor, Applicant collected patient sex, ethnicity, body mass index, and age from the electronic health record and stored this information in an encrypted database.

[0090] Example 1,4, Radiographic Data Retrieval

[0091] Preoperative MRIs of each analyzed tumor were retrospectively accessed through the electronic health record and annotated by expert board-certified neuroradiologists. Features were extracted from the radiological report into a standardized sheet of semantic variables. All radiographic features outside of 3D maximum tumor dimensions were binarized and stored in a database separate from the corresponding molecular classifications in a blinded fashion to reduce potential sources of bias.

[0092] Example 1.5. Radiomic Data Retrieval

[0093] Preoperative T1 post-contrast and T2 FLAIR MRIs were anonymized and exported as DICOMs and uploaded to 3D Slicer. Each tumor was semi-automatically segmented by an expert user and manually refined using the pyRadiomics plugin in 3D slicer. Applicant performed normalization and resampling of the images for homogenous comparison across scans of different slice thicknesses and acquisition patterns. Tables 1-3 provide listings of the extracted features.SkewnessAuto correlationJoint AverageCluster ShadeContrastDifference AverageDifference EntropyDifference VarianceJoint EnergyJoint EntropyIdmIdmnIdIdnInverse VarianceMaximum ProbabilitySum AverageHigh Gray Level Run EmphasisLong Run EmphasisLong Run High Gray Level EmphasisLow Gray Level Run EmphasisLong Run Low Gray Level EmphasisRun PercentageRun Length Non Uniformity NormalizedRun VarianceShort Run EmphasisShort Run High Gray Level EmphasisShort Run Low Gray Level EmphasisShort Run High Gray Level EmphasisHigh Gray Level Zone EmphasisGlszm VarianceLow Gray Level Zone EmphasisSmall Area High Gray Level EmphasisDependence Non Uniformity NormalizedDependence VarianceHigh Gray Level EmphasisLarge Dependence EmphasisLarge Dependence High Gray Level EmphasisLarge Dependence Low Gray Level EmphasisLow Gray Level EmphasisSmall Dependence High Gray Level EmphasisTable 1. Non-trending extracted features.RadiograhicRadiomicVolume ElongationTV Least Axis LengthAP Maximum 2D Diameter ColumnSI Maximum 2D Diameter RowMaximum Dimension Maximum 2D Diameter SliceSide Maximum 3D DiameterLocation Group Mesh VolumeMultiple Lesions Minor Axis LengthGraded Edema SphericityAdjacent Bony Changes Surface AreaAdjacent Bony Changes Qualitative Surface Volume RatioMRI Enhancement Pattern Voxel VolumeCystic Ten percentCentral Necrosis or Rim Enhancement EnergyApparent Diffusion Coefficient EntropyEdema Positive Interquartile RangeMean Absolute DeviationMinimumRobust Mean Absolute DeviationTotal EnergyUniformityVarianceCluster ProminenceClusterT endencyImclSum EntropySum SquaresGlrlm Non UniformityGlrlm Non Uniformity NormalizedGlrlm Level VarianceRunRun Length Non UniformityGlszm Non UniformityGray Level Non Uniformity NormalizedLarge Area EmphasisLarge Area High Gray Level EmphasisLarge Area Low Gray Level EmphasisSize Zone Non UniformitySize Zone Non Uniformity NormalizedSmall Area EmphasisSmall Area Low Gray Level EmphasisZone EntropyZone PercentageZone V arianceDependence Non UniformityGray Level Non UniformityGray Level VarianceSmall Dependence EmphasisSmall Dependence Low Gray Level EmphasisTable 2. Non-selected extracted features.Pre-MRI Radiograhic RadiomicSex Intensity on T1 non-contrast Dependence Entropy Tumor Margins Major Axis Length Supra vs. Infra Correlation Location FlatnessImc2Table 3. Selected extracted features.

[0094] Example 1.6. Statistical Analysis

[0095] All statistical analyses and modeling were performed using R. Summary statistics were calculated for all radiographic variables to compare tumors that were classified as either benign (MenG A / B) or aggressive (MenG C) subgroups. Applicant decided to group MenG A and MenG B togetherfor analysis, as they are both benign. As such, they are more clinically meaningful to preoperatively prognosticate together than apart, in contrast with aggressive MenG C tumors, which invariably recur. Categorical variables were analyzed using Pearson’s Chi-squared or Fischer’s Exact test, and continuous variables were analyzed using the Wilcoxon rank-sum test. Non-ordinal qualitative variables were binarized.

[0096] After performing feature selection, Applicant trained two 10-repeat 10-fold cross-validated models using all tumors resected from 2012-2020 (training set). One model was based on a random forest algorithm; the other was based on a single-layered neural network. Subsequently, Applicant tested these models on tumors collected from 2021-2022 (validation set). Applicant evaluated all models using Area Under the Receiver Operating characteristic Curve (AUROC).

[0097] Example 1,7, Results

[0098] To determine whether MRI features are sufficient to identify aggressive meningiomas according to Applicant’s molecular classification, Applicant analyzed pre-MRI variables, MRI variables and a combination of both using two patient cohorts, dating back over a decade. Applicant then used two computational model approaches: a neural network algorithm and a random forest algorithm with more tunable hyperparameters to maximize AUROC.

[0099] First, Applicant asked if clinical data obtained from 178 patients prior to imaging was sufficiently sensitive and specific to identify aggressive tumors. Using only clinical variables such as sex, ethnicity, and age, Applicant was unable to adequately differentiate between MenG A / B (benign) and MenG C (aggressive) meningiomas (AUROC=0-59, FIGS. 2A-2B). Because previous literature has suggested that peritumoral edema is associated with the tendency of meningioma to behave invasivcly, Applicant aimed to see if adding edema to the prc-MRI features would be predictive. This was not the case (FIG. 2C).

[0100] Next, Applicant asked if including preoperative MRI features could better predict aggressiveness. Applicant annotated the same 178 tumors radiographically, performed tumor segmentation manually on two different MRI sequences (T1 post-contrast, T2 FLAIR), and extracted radiomic features of tumors, attempting to predict their molecular classification. Applicant collected a total of 133 features, including pre-MRI variables, MRI-extracted radiographic, and radiomic variables (FIG. 2D).

[0101] Fifty-eight features including age (p=0.78), BMI (p=0.56), ethnicity (p=0.55), white vs. non- white ethnicity (p=0.18), and all T2 FLAIR radiomic variables were excluded as noncontributory. The remaining 75 features were found to vary with benign vs. aggressive status. Applicant further eliminated variables that were potentially redundant or unhelpful in identifying MenG status using a recursive feature elimination algorithm (FIG. 2D).

[0102] Ultimately, a combination of 10 variables yielded the best performance (FIGS. 3A- 3J). None of these ten variables showed good discriminatory ability (AUROOO.8) in isolation, but the highest AUROC was associated with supratentorial location.

[0103] Applicant next turned to the computational approaches with these ten variables. Using internal cross-validation on the 2012-2020 training set of tumors, the neural network algorithm achieved a maximum AUROC of 0.84 and the random forest algorithm of a AUROC of 0.85 (FIG. 4A). At their respective optimal thresholds, the neural network model had an 80% accuracy, while the random forest model had a 79% accuracy. Thus, including preoperative MRI variables drastically improved outcome prediction.

[0104] Having developed two models with high sensitivity and specificity for identifying aggressive meningiomas, Applicant next assessed the generalizability of the approach using radiographically annotated tumors diagnosed between 2021-2022 by testing the models on this new tumor set (n=66, FIG. 4B). The neural network model achieved an AUROC of 0.88 and the random forest model an AUROC of 0.89. At optimal thresholds, the neural network achieved an accuracy of 83%, while the random forest model achieved an accuracy of 89%. Notably, these thresholds remained consistent with the thresholds identified in the previous models. This demonstrates that the models previously built arc not overfit to their training data and can effectively differentiate MenG C (aggressive) tumors.

[0105] Given that both training and validation sets were highly consistent, Applicant used all available tumors (n=244, FIG. 4C) to retrain the neural network and random forest models to determine whether the AUROC could be further improved with an increased sample size. Both algorithms produced models with an AUROC of 0.86. At their respective optimal thresholds, the neural network model and random forest models have accuracies of 82%. AUROC values and accuracies were stable between all three sets, strongly suggesting the sample size is sufficient and the algorithms can reliably predict aggressive tumors preoperatively.

[0106] MRI-based models that reliably predict meningioma aggressiveness would alter clinical practice by enabling the medical team to plan surgery and post-operative care / treatment knowing whether the tumor is benign or malignant. Applicant therefore compiled a list of features commonly used to evaluate meningioma severity pre-operatively (FIG. 5). Surprisingly, many of these features including volume, tumor necrosis, rim enhancement, restricted vs. non-restricted Apparent Diffusion Coefficient (ADC), graded edema, and binary edema were excluded as individual variables because they were already incorporated into others features of the final set (FIG. 5). For example, edema is included in Dependence Entropy, among other variables (FIG. 5).

[0107] Applicant also carefully reviewed the mistakes made by the two models to further improve the decision making process. Most of the errors — 9 out of 11 (82%, p=0.036) made by the neural network model and 7 out of 7 (100%, p=0.013) made by the random forest model — were in the direction of underdiagnosis with tumors with edema on preoperative MRI. Patients with edema should be considered likely to have aggressive meningioma (FIGS. 6A-6D), and the surgical team prepared accordingly, given previous literature indicating that peritumoral edema is associated with invasive behavior. In patients presenting without edema, these models, especially the random forest model, are extremely accurate and justify the clinical utility of radiomics for meningioma. Taken together, these results demonstrate that the MRI-based models can preoperatively predict malignant MenG C meningiomas.

[0108] Example 1.8. Discussion

[0109] Using one of the largest cohorts of molecularly defined meningiomas to date, Applicant has identified radiographic and radiomic features that predict aggressive (MenG C) tumors with robust sensitivity and specificity. These results have profound implications for planning treatment approaches to patients.

[0110] The features selected in this Example cannot be treated in an additive fashion. For example, large size, parafalcine location and male sex do not mean the tumor is aggressive while a tumor of large size, parafalcine location, and female sex is not. Nevertheless, the ten variables shed light on several interesting aspects of meningioma pathophysiology. Previous studies have likewise associated convexity or parafalcine location with higher WHO grades and specific mutational landscapes, particularly in the NF2 gene. Similarly, indistinct tumor margins are known to suggest pial invasion and a difficult, more risky surgery and are associated with MenG C.

[0111] Applicant also found that hypointensity on T1 non-contrast imaging was associated with a more aggressive outcome. While hypointensity has been postulated to be secondary to axonal reduction and extracellular edema, previous literature examining varying intensities on T1 and T2 noncontrast imaging has failed to demonstrate a correlation with WHO grade or proliferative activity (i.e., Ki-67). Small sample size and reliance on WHO grading may have obscured the statistical association of this radiographic finding.

[0112] Exclusions from Applicant’s final set of predictive features included volume, homogeneous vs. heterogeneous MRI enhancement pattern, and central necrosis / rim enhancement. Applicant’ s findings agree with previous literature that all these features could be helpful in identifying aggressive meningiomas, but the statistically significant correlations between a majority of the selected features suggest they provide redundant information. A surprising exclusion from the final features was ADC (apparent diffusion coefficient), a promising prognostic factor identified by the emerging field of radiogenomics in meningioma studies. A prior study found that a high ADC signal correlates with prognosis and a further study showed that high diffusion restriction on ADC predicts TERT promoter mutation status in Grade 2 meningiomas. Applicant’s feature selection process determined that adding ADC to the models or replacing extant features with ADC did not improve model performance, but 8 / 10 identified features had statistically significant correlations with it (FIG. 5). This indicates that any information ADC provides in differentiating benign from aggressive tumors is redundant and / or less informative than the ten features Applicant selected.

[0113] Another feature that was absent from the final 10 variables was edema, which was positively correlated with “Major Axis Length” (FIG. 5). Given that the random forest model made no errors in patients presenting without edema, this model could be particularly useful for patients who want to delay surgery or forgo surgery because they are asymptomatic, high-risk for surgery, or elderly (FIG. 5).

[0114] Applicant envisions at least two scenarios in which the use of the models would benefit not only the patient but also the healthcare system. In the first, patients predicted to have an indolent tumor (model prediction is benign and patient does not have cerebral edema) can more comfortably extend the time interval between imaging and might be able to postpone surgery. In the second scenario, a prediction of an aggressive tumor prior to surgery or the presence of edema would promptthe surgeon to work more aggressively or plan more stringent post-operative care, such as more frequent imaging.

[0115] In sum, molecular classification in meningioma can be predicted by a combination of clinical, radiographic and radiomic features. Identifying aggressive tumors prior to surgery can alter surgical planning and postoperative care for a sizable proportion of patients.

[0116] Without further elaboration, it is believed that one skilled in the art can, using the description herein, utilize the present disclosure to its fullest extent. The embodiments described herein are to be construed as illustrative and not as constraining the remainder of the disclosure in any way whatsoever. While the embodiments have been shown and described, many variations and modifications thereof can be made by one skilled in the art without departing from the spirit and teachings of the invention. Accordingly, the scope of protection is not limited by the description set out above, but is only limited by the claims, including all equivalents of the subject matter of the claims. The disclosures of all patents, patent applications and publications cited herein are hereby incorporated herein by reference, to the extent that they provide procedural or other details consistent with and supplementary to those set forth herein.

Claims

CLAIMS1. A method of predicting a molecular subtype of a tumor of a subject, said method comprising: receiving preoperative imaging data of the tumor; extracting one or more features of the imaging data; and correlating the one or more extracted features of the imaging data to a molecular subtype of the tumor.

2. The method of claim 1, wherein the tumor comprises a brain tumor.

3. The method of claim 1, wherein the tumor comprises meningioma.

4. The method of claim 1, wherein the subject is a human being.

5. The method of claim 1, wherein the preoperative imaging data comprises radiological imaging data, computed tomography (CT) scan data, positron emission tomography (PET) data, magnetic resonance imaging (MRI) data, age of the subject, gender of the subject, location of the subject, or combinations thereof.

6. The method of claim 1, wherein the preoperative imaging data comprises magnetic resonance imaging (MRI) data.

7. The method of claim 1, wherein the extraction comprises semi-automatic segmentation of the imaging data.

8. The method of claim 1, wherein the extracted features of the imaging data are selected from the group consisting of radiomic features, radiographic features, first order features, quantitative features, qualitative features, information related to tumor location, information related to tumor length, subject’s gender, information related to tumor volume, information related to tumor edema, supra- vs. infratentorial status of the tumor, presence of multiple tumors, local bony changes, enhancement pattern of the tumor, heterogenous enhancement pattern of the tumor, cystic degeneration of the tumor, central necrosis of the tumor, presence of intra-tumoral flow voids, tumor intensity on T1 / T2 sequences, tumor restriction on apparent diffusion coefficient (ADC) sequence, tumor shape, distinct tumor margins on T2 (“CSF cleft”), tumor hypointensity on T1 non-contrast sequences, indistinct tumor margins, lmc2, dependence entropy, correlation, appearance of dural tail on the tumor, susceptibility changes of the tumor on gradient or fast field echo sequences, or combinations thereof.

9. The method of claim 1, wherein the extracted features of the imaging data are selected from the group consisting of information related to tumor location, information related to tumor length, subject’s gender, supra- vs. infratentorial status of the tumor, tumor shape, tumor hypointensity on T1 non-contrast sequences, indistinct tumor margins, lmc2, dependence entropy, correlation, or combinations thereof.

10. The method of claim 1, wherein the extracted features comprise information related to tumor edema.

11. The method of claim 1 , wherein the correlation comprises correlation to a benign or malignant subtype of the tumor.

12. The method of claim 1, wherein the correlation comprises correlation to a benign or malignant molecular subtype of meningioma.

13. The method of claim 12, wherein the correlation comprises classification of extracted imaging data into a molecular subtype of meningioma group A (MenG A), meningioma group B (MenG B), meningioma group C (MenG C), or combinations thereof.

14. The method of claim 1, wherein the correlation occurs through the utilization of a machinelearning algorithm trained on the extracted features of the image data.

15. The method of claim 1, further comprising a step of outputting the correlated molecular subtype of the tumor.

16. The method of claim 1, further comprising a step of implementing a treatment decision.

17. The method of claim 16, wherein the treatment decision comprises monitoring the course of the tumor, implementing a tumor treatment regimen, implementing a tumor management regimen, or combinations thereof.

18. The method of claim 16, wherein the treatment decision comprises implementing a tumor treatment regimen, wherein the tumor treatment regimen comprises administering a therapeutic agent to the subject, removing the tumor, or combinations thereof.

19. The method of claim 1, wherein the method is used to prognosticate the aggressiveness of the tumor, and wherein the method comprises:(a) receiving on at least one processor, at least one preoperative image from a subject with a brain tumor;(b) performing semi-automatic segmentation of the preoperative image;(c) classifying the segmented image into at least two groups using at least one machine learning based classifier; and(d) outputting, using at least one processor, a classification of the sample concerning prognosis of the tumor in the subject based on steps (a)-(c).

20. The method of claim 19, where the preoperative image is a magnetic resonance image (MRI).

21. The method of claim 19, wherein the semi-automatic segmentation comprises extraction of first order features.

22. The method of claim 19, wherein the extracted data comprises radiomic features, radiographic features, or combinations thereof.

23. The method of claim 19, where the tumor is a meningioma in a human brain.

24. The method of claim 23, wherein the classification comprises a molecular subtype classification as a MenG A, MenG B, MenG C, or a combination thereof.

25. A computing device for predicting a molecular subtype of a tumor of a subject, wherein the computing device comprises one or more computer readable storage mediums having a program code embodied therewith, wherein the program code comprises: programming instructions for receiving preoperative imaging data of the tumor; programming instructions for extracting one or more features of the imaging data; and programming instructions for correlating the one or more extracted features of the imaging data to a molecular subtype of the tumor.

26. The computing device of claim 25, further comprising programming instructions for recommending a treatment decision.

27. The computing device of claim 25, wherein the computing device is a component of a diagnostic test for predicting a molecular subtype of a tumor of a subject.

28. The computing device of claim 25, wherein the tumor comprises meningioma.

29. The computing device of claim 25, wherein the prcopcrativc imaging data comprises radiological imaging data, computed tomography (CT) scan data, positron emission tomography (PET) data, magnetic resonance imaging (MRI) data, age of the subject, gender of the subject, location of the subject, or combinations thereof.

30. The computing device of claim 25, wherein the preoperative imaging data comprises magnetic resonance imaging (MRI) data.

31. The computing device of claim 25, wherein the extracted features of the imaging data are selected from the group consisting of radiomic features, radiographic features, first order features, quantitative features, qualitative features, information related to tumor location, information related to tumor length, subject’s gender, information related to tumor volume, information related to tumor edema, supra- vs. infratentorial status of the tumor, presence of multiple tumors, local bony changes, enhancement pattern of the tumor, heterogenous enhancement pattern of the tumor, cystic degeneration of the tumor, central necrosis of the tumor, presence of intra-tumoral flow voids, tumor intensity on T1 / T2 sequences, tumor restriction on apparent diffusion coefficient (ADC) sequence, tumor shape, distinct tumor margins on T2 (“CSF cleft”), tumor hypointensity on T1 non-contrast sequences, indistinct tumor margins, lmc2, dependence entropy, correlation, appearance of dural tail on the tumor, susceptibility changes of the tumor on gradient or fast field echo sequences, or combinations thereof.

32. The computing device of claim 25, wherein the extracted features of the imaging data are selected from the group consisting of information related to tumor location, information related to tumor length, subject’s gender, supra- vs. infratentorial status of the tumor, tumor shape, tumor hypointensity on T1 non-contrast sequences, indistinct tumor margins, lmc2, dependence entropy, correlation, or combinations thereof.

33. The computing device of claim 25, wherein the programming instructions for correlation comprise correlation to a benign or malignant molecular subtype of meningioma.

34. The computing device of claim 33, wherein the correlation comprises classification of extracted imaging data into a molecular subtype of meningioma group A (MenG A), meningioma group B (MenG B), meningioma group C (MenG C), or combinations thereof.

Images