Systems and methods for determining memory metrics

Abstract

Embodiments of the present disclosure provide methods, systems and non-transitory computer-readable media of determining memory metrics. An example method includes: presenting a plurality of stimuli to a patient; collecting behavioral responses from the patient to the plurality of stimuli; extracting signature, using at least one mathematical model, from the behavioral responses related to memory processes; estimating, using a first computational model, a performance metric for each particular stimulus based on the extracted signature, the performance metric relates to speed of forgetting; and estimating, using a second computational model, memory metric of the patient based on the performance metrics of the stimuli.

Description

SYSTEMS AND METHODS FOR DETERMINING MEMORY METRICS

[0001] This application claims priority to U.S. Provisional Application 63 / 579,843, filed August 31, 2023, which application is hereby incorporated by reference, in its entirety, for any purpose.TECHNICAL FIELD

[0002] Examples described herein relate to human memory, including systems and methods for determining memory metrics. Examples include the use of computational models to determine memory metrics.BACKGROUND

[0003] Long-term memory, that is, the mental faculty to store and remember information over extended amounts of time (from seconds to years), is necessary to carry out a normal, functional, and fulfilled existence.

[0004] It is generally agreed that memories tend to fade over time if not actively repeated or rehearsed. The speed at which such fading occurs determines the degree of forgetting, and the cognitive impairment described as amnesia is characterized by abnormally fast forgetting.

[0005] As each individual ages, the long-term memory function of each individual naturally declines and such individual can be impacted by age-related neuropsychological diseases, for example, Parkinson’s disease, Alzheimer’s disease “AD,” Fronto-Temporal Degeneration, and Pick’s disease. With the aging population, the number of people experiencing memory disorders is expected to increase. There are also examples of non-age-related impact on longterm memory, caused by neuropsychological and other disorders, dysfunctions, or states such as depression, clinical migraine, post-anesthesia or post-cancer treatment, or mild-traumatic brain-injury that might increase forgetting rate. Because long-term memory plays a critical role in daily activities, individuals with memory disorders need extensive care and supervision. Thus, societal cost of care and supervision associated with long-term memory disorders is expected to inflate.

[0006] In order for accurate and efficient care to reduce such cost, fast and accurate diagnosis of memory disorders or dysfunctions (e.g., after a concussion, during phases of depression, clinical migraines, or burn-out) is the first step. Despite significant efforts to understand and treat these conditions, progress has been slow. One major challenge is the lack of understanding of the relationship between long-term memory decline and the underlying neuropathology. To gain a better understanding of this relationship, it is beneficial to haveprecise and longitudinal assessments of memory function that can identify which aspect of long-term memory is declining, at what rate, and due to which underlying affected neural circuit. However, current clinical tools for memory assessments are not adequate for this purpose. Conventionally, memory disorders may be established through standardized assessment tests. The standardized assessment tests are subject to a number of problems. The tests are complex to develop, requiring extensive research and reference data collection. Existing assessment tests are limited to standardized questionnaires. Test content is fixed and cannot be repeated, because the repetition has been shown to affect results. Translation and standardization are necessary for different languages, each with its own norms (e.g., culture, etc.). Expert professionals are required to administer the tests due to their specialist nature. Lastly, financial and societal barriers hinder widespread population screening using these tests.SUMMARY

[0007] Examples described herein are directed towards systems, methods and computer- readable media for determining memory metrics. An example method includes presenting a plurality of stimuli to a patient; collecting behavioral responses from the patient to the plurality of stimuli; extracting signature, using at least one mathematical model, from the behavioral responses related to memory processes; estimating, using a first computational model, a performance metric for each particular stimulus based on the extracted signature, the performance metric relates to speed of forgetting; and estimating, using a second computational model, a memory metric of the patient based on the performance metrics of the stimuli.

[0008] In some examples, extracting the signature includes removing at least one portion of the behavioral responses due to non-memory processes. The at least one portion of the behavioral responses may be related to a perception retrieval. The at least one portion of the behavioral responses may be related to a motor response.

[0009] In some examples, adjusting a next stimulus of the plurality of stimuli based on the memory metric. In some examples, adjusting the next stimulus of the plurality of stimuli may be to increase an information gain in updating the memory metric associated with memory capacity of the patient. In some examples, adjusting the next stimulus of the plurality of stimuli may include at least one of adjusting a difficulty level of a task including the plurality of stimuli or a speed of presenting the next stimulus.

[0010] In some examples, the method may further include displaying diagnostic-related information.

[0011] An example system includes at least one processor and one or more memory devices. The one or more memory devices storing instructions that, when executed by the processor, configure the system to: present a plurality of stimuli to a patient; collect behavioral responses from the patient to the plurality of stimuli; extract a signature related to memory processes, using at least one mathematical model, from the behavioral responses; estimate, using a first computational model, a performance metric for each particular stimulus based on the extracted signature, the performance metric relating to speed of forgetting; and estimate, using a second computational model, of a memory metric of the patient based on the performance metrics of the plurality of stimuli.

[0012] In some examples, the extract the signature includes remove at least one portion of the behavioral responses due to non-memory processes. The at least one portion of the behavioral responses may be related to a perception retrieval. The at least one portion of the behavioral responses may be related to a motor response.

[0013] In some examples, the instructions further configure the system to adjust a next stimulus of the plurality of stimuli based on the memory metric. In some examples, said adjust the next stimulus of the plurality of stimuli may be to increase an information gain in updating the memory metric associated with memory capacity of the patient. In some examples, said adjust the next stimulus of the plurality of stimuli may include at least one of adjusting a difficulty level of a task including the plurality of stimuli or a speed of presenting the next stimulus.

[0014] In some examples, the instructions further configure the system to display diagnostic- related information.

[0015] An example non-transitory computer-readable medium including instructions that when executed by a computer cause the computer to: present a plurality of stimuli to a patient; collect behavioral responses from the patient to the plurality of stimuli; extract a signature related to memory processes, using at least one mathematical model, from the behavioral responses; estimate, using a first computational model, a performance metric for each particular stimulus based on the extracted signature, the performance metric relating to speed of forgetting; and estimate, using a second computational model, of a memory metric of the patient based on the performance metrics of the plurality of stimuli.

[0016] In some examples, the extract the signature includes remove at least one portion of the behavioral responses due to non-memory processes. The at least one portion of the behavioral responses may be related to a perception retrieval. The at least one portion of the behavioral responses may be related to a motor response.

[0017] In some examples, the instructions further configure the computer to adjust a next stimulus of the plurality of stimuli based on the memory metric. In some examples, said adjust the next stimulus of the plurality of stimuli is to increase an information gain in updating the memory metric associated with memory capacity of the patient. In some examples, said adjust the next stimulus of the plurality of stimuli comprises at least one of adjusting a difficulty level of a task including the plurality of stimuli or a speed of presenting the next stimulus.

[0018] In some examples, the instructions further configure the computer to display diagnostic-related information.BRIEF DESCRIPTION OF THE DRAWINGS

[0019] FIG. 1 is a schematic diagram of a system for determining memory metrics in accordance with examples described herein.

[0020] FIG. 2 is a block diagram of an example system for determining memory metrics in accordance with examples described herein.

[0021] FIG. 3 shows an equation representing log odds of retrieving a memory in accordance with examples described herein.

[0022] FIG. 4 shows an equation representing a trace-specific decay rate in accordance with examples described herein.

[0023] FIG. 5 shows an equation representing a probability of a correct response to a stimulus related to the activation in accordance with examples described herein.

[0024] FIG. 6 shows an equation representing a response time associated with a correct response to a stimulus related to the activation in accordance with examples described herein.

[0025] FIG. 7 shows a diagram of predicted forgetting associated with a specific value of the memory metric in accordance with examples described herein.

[0026] FIG. 8 shows examples of a software interface in accordance with examples described herein.

[0027] FIG. 9 includes diagrams showing a distribution of speed of forgetting (SoF) values across all lessons and patients in accordance with examples described herein.

[0028] FIG. 10 is a diagram showing correlations between memory metrics measured across lessons of different materials.

[0029] FIG. 11 includes diagrams showing a distribution of memory metrics across patients affected by mild cognitive impairment (MCI) and age-matched healthy controls in accordance with examples described herein.

[0030] FIG. 12A shows receiver operating characteristic (ROC) classification performances for SoF and for response accuracy from a single session in accordance with examples described herein.

[0031] FIG. 12B shows average ROC classification performances for SoF and for response accuracy across sessions in accordance with examples described herein.

[0032] FIG. 13 shows probability of MCI by different levels of memory metrics in accordance with examples described herein.

[0033] FIG. 14 shows age-related changes in memory metric over time in MCI patients and healthy controls in accordance with examples described herein.

[0034] FIG. 15 shows a distribution of memory metric values across different groups and types of stimuli in accordance with examples described herein.

[0035] FIG. 16 shows a detailed view of memory metrics across materials and patient subgroups in accordance with examples described herein.DETAILED DESCRIPTION

[0036] The following description of certain embodiments is merely exemplary in nature and is in no way intended to limit the scope of the disclosure or its applications or uses. In the following detailed description of embodiments of the present systems, devices, and methods, reference is made to the accompanying drawings which form a part hereof, and which are shown by way of illustration specific to embodiments in which the described systems, methods and computer-readable media may be practiced. In some instances, well-known neurological anatomy, clinical procedures, circuits, control signals, timing protocols, and / or software operations have not been shown in detail in order to avoid unnecessarily obscuring the described embodiments. It is to be understood that other embodiments may be utilized and that structural and logical changes may be made, without departing from the spirit and scope of the disclosure. The following detailed description is therefore not to be taken in a limiting sense for the appended claims.

[0037] Methods to assess a long-term memory function of an individual (e.g., a subject, a patient, etc.) are provided. Example methods may assess memory function of the individual using a memory metric. The memory metric may be a quantitative and / or precise measure of memory function. Example methods include collecting data through an automated digital session with a subject. Examples of the digital session may not utilize any human supervision. Each session may be administered remotely on affordable digital devices (e.g., desktop computers, laptop computers, mobile smartphones, wearable devices, and / or tablets). In mostexamples, the digital device used by the subject to perform the automated digital session may be connected with other electronic devices via a peer-to-peer connection or a network (e.g., the Internet, Wi-Fi, a local area network “LAN,” a wide area network “WAN,” Bluetooth “BT,” near-field communication “NFC,” etc.) in constant or intermittent communication.

[0038] Example methods use a collection of responses from a subject to a series of stimuli. The responses may include, for example, inputs to an electronic device, such as keystrokes, mouse clicks, spoken responses, and / or touchscreen gestures from the subject and corresponding times elapsed for providing such inputs (e.g., response time). These responses are collected for a subject throughout a session. In some examples, a session may include an open-ended, continuous series of responses by a subject to a series of corresponding experimental stimuli presented over a software interface (e.g., displayed on a digital device, sound produced from the digital device, etc.).

[0039] In some examples, the digital device may transmit the collected responses to a computer device. The computer device may add the collected responses to temporary record of all responses through the session in association with the presented stimuli.

[0040] The stimuli may be, for example, questions regarding a meaning of one or more words, a prompt to identify one or more locations on a map, or a prompt for a recall of factual information. During a session, a particular stimulus (e.g., a particular question) may be presented once or multiple times.

[0041] After each response, the recorded responses are examined by a computer program executed on a computer. The computer program uses these responses to estimate a performance metric. The performance metric may include, for example, a speed of forgetting (SoF) for each stimulus. For example, multiple responses to the same stimulus over time in a session may be used to estimate a performance metric associated with the stimulus. For example, during the session the subject may initially provide an incorrect response to the stimulus, but over time the subject may begin to provide a correct response. The times at which responses are provided may be related to the performance metric (e.g., the SoF) for the stimulus.

[0042] Following each response, a current estimated value of SoF of the subject for the current stimulus may be adjusted upwards or downwards to reflect a performance of the subject. If the response is correct and fast, the SoF may be revised downwards. Conversely, if the response is slow and / or incorrect, the SoF may be revised upwards. A memory metric of a subject may be provided by combining the SoFs for all stimuli in a session.

[0043] Systems, methods and computer-readable media for determining memory metrics using one or more computational cognitive models may be advantageous as they may reasonably represent a relationship between long-term memory decline and neuropathology. Examples described herein employ a computational cognitive model that simulates encoding and passive forgetting based on established cognitive and biological principles, providing a framework for understanding the underlying mechanisms of memory decline in aging and neurodegenerative conditions. Therefore, the memory metric determination technique may be used for predictions about the progression of memory declination and / or identification of potential therapeutic targets.

[0044] FIG. 1 is a schematic diagram of a system 100 for determining memory metrics in accordance with examples described herein. The system 100 includes a computing device 102 and a digital device 130. In some examples, the digital device 130 may be coupled to the computing device 102. In some examples, the digital device 130 and the computing device 102 may be an integrated device.

[0045] The digital device 130 may function as an interface to a subject 150. The digital device 130 may be implemented, for example, using one or more smartphones, cell phones, tablets, medical devices, wearable devices, computers such as laptop computers, desktop computers, servers, automobiles, appliances, or other electronic devices. Examples of digital devices described herein may include components, such as one or more processors 136, one or more memory devices 138, and communication interface 148. Examples of processors, such as processor 136, may be implemented using one or more central processing units (CPUs), graphics processing units (GPUs), field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), controllers, microcontrollers, or other circuitry. Processors described herein may be used to implement software and / or firmware systems. For example, the processor 136 may be used to execute one or more executable instructions encoded on computer-readable media (e.g., one or more memory devices 138). While a single processor 136 is shown in FIG. 1, it is to be understood that any number of processors may be used, and multiple processors may be in communication with each other to perform functions of a processor described herein.

[0046] Examples of computing devices described herein may include memory, such as the memory device 138 of FIG. 1. Generally, any number or kind of memory may be used including random access memory (RAM), read only memory (ROM), solid state drive (SSD), hard disk drives (HDDs), flash memory, etc., or other computer-readable media. While the single memory device 138 is shown in FIG. 1, it is to be understood that any number of memory devices may be used, and the executable instructions and / or data may be distributedacross multiple memories accessible to the processor 136. The memory devices 138 may include data memory 140 and program memory 142. The program memory 142 may be a non-transitory computer-readable medium storing computer-executable instructions, including executable instructions for presenting stimuli 144 and executable instructions for collecting responses 146. In some examples, the program memory 142 may be encoded with instructions which, when executed, may cause the digital device 130 to provide stimuli and receive responses, such as behavioral responses related to memory processes, as described herein. In some examples, the data to be stored in the data memory 140 may include, for example, data for performing instructions encoded in the program memory 142. In some examples, the data may include a set of stimuli to be presented, such as images, sounds, and / or strings. In some examples, the set of stimuli may be provided from the computing device 102. In some examples, the set of stimuli may be stored in the data memory 140 associated with a set of corresponding identifiers, such as numbers, characters, or a combination thereof. In some examples, the data may include sets of response and response time for providing the corresponding response provided by each subject. Each set of response and response time is further stored in association with a presented stimulus or an identifier of the presented stimulus that caused the related response.

[0047] Examples of digital devices described herein may include stimuli presentation circuitry, such as a display 132. In some examples, the executable instructions for presenting stimuli 144 performed by the processor 136 may cause the stimuli presentation circuitry, such as the display 132, to present the stimuli. In some examples, the display 132 may be integrated into the digital device 130. In some examples, the display 132 may be external to the digital device 130 and coupled to the digital device 130 via one or more wires (e.g., a data cable, such as a universal serial bus “USB” cable, an unshielded twisted pair “UTP” cable, or a video cable such as a Hi-definition multimedia interface “HDMI” cable, a DisplayPort “DP” cable, or other standardized or proprietary cables) or wirelessly (e.g., the Internet, Wi-Fi, LAN, WAN, BT, NFC, or other standardized or proprietary wireless communication). In some examples, the stimuli may be provided to the subject 150 through another type of output devices, such as a sound producing device (e.g., speaker, headphone, etc.).

[0048] Examples of digital devices described herein may include data acquisition circuitry, such as an input interface 134. In some examples, the executable instructions for collecting responses 146 performed by the processor 136 may cause the input interface 134 to collect the responses. The input interface 134 may receive one or more responses through one or more entries of the subject 150. In some examples, the input interface 134 may receive an entry of the subject 150. In some examples, the one or more input interfaces 134 may includean electronic device (e.g., keys, buttons, keyboard, mouse, touchscreen, microphone, etc.) integrated into the digital device 130 placed in proximity to the subject 150. In some examples, the display 132 and one or more input interfaces 134 may be integrated as one device (e.g., a touchscreen). In some examples, the one or more input interfaces 134 may provide a communication interface with an electronic peripheral device (e.g., keyboard, mouse, touchscreen, microphone) that may be coupled to the digital device 130 via a cable or wirelessly. In some examples, the responses provided by the subject 150 using the input interface 134 may be, for example, in the form of keystrokes, mouse clicks, spoken responses, and / or touchscreen gestures, etc. In some examples, each behavioral response may correspond to each preceding stimulus; thus each stimulus and each behavioral response may be paired. The digital device 130 may store the collected responses in the data memory 140, and provide the collected responses to the computing device 102 from the communication interface 148.

[0049] The computing device 102 may be implemented, for example, using one or more smartphones, cell phones, tablets, medical devices, wearable devices, computers such as laptop computers, desktop computers or servers, automobiles, appliances, or other electronic devices. Examples of computing devices described herein may include components, such as one or more processors 104, one or more memory devices 106, a communication interface 128, a display 126, and an internal bus 124. Examples of computing devices described herein may generally include one or more processors, such as a processor 104 of FIG. 1. Processors, such as one or more processors 104, may be implemented using one or more CPUs, GPUs, FPGAs, ASICs, controllers, microcontrollers, or other circuitry. Processors described herein may be used to implement software and / or firmware systems. While a single processor 104 is shown in FIG. 1, it is to be understood that any number of processors may be used, and multiple processors may be in communication with each other to perform functions of a processor described herein. The processor 104 may be coupled to (e.g., in communication with) a memory device 106, a communication interface 128, and / or one or more displays 126 of FIG. 1 via an internal bus 124. In some examples, the processor 104 may be used to execute one or more executable instructions encoded on computer-readable media (e.g., memory).

[0050] Examples of computing devices described herein may include memory, such as the memory device 106 of FIG. 1. Generally, any number or kind of memory may be used including ROM, RAM, flash memory, one or more SSDs, one or more HDDs, or other computer-readable media. While the single memory device 106 is shown in FIG. 1, it is to be understood that any number of memory devices may be used, and the executable instructions and / or data may be distributed across multiple memories accessible to the processor 104. The memory device 106 may include program memory 110 and data memory 108. In someexamples, the program memory 110 and the data memory 108 may be implemented as separate segments of the memory device 106 as one or more integrated memory devices. In some examples, the program memory 110 and the data memory 108 may be implemented as separate memory devices of the same kind or different kinds. In some examples, any of the program memory 110 and / or the data memory 108 may be fixed in the computing device 102. In some examples, any of the program memory 110 and / or the data memory 108 may be attachable to and detachable from the computing device 102. The program memory 110 may be implemented as a non-transitory computer-readable medium storing computer-executable instructions, including executable instructions for determining memory metrics 112. The executable instructions for determining memory metrics 112 may include executable instructions for receiving and storing responses 114, executable instructions for extracting signature from responses 116, executable instructions for estimating a performance metric related to speed of forgetting 118, executable instructions for estimating a memory metric based on performance metrics 120 and / or executable instructions for determining stimulus 122.

[0051] The data memory 108 described herein may store data. In some examples, the data to be stored in the data memory 108 may include, for example, data for performing instructions encoded in the program memory 110. In some examples, the data stored in the data memory 108 may include data to be exchanged with external devices, such as the digital device 130. For example, the data to be exchanged may include information about a stimulus to be presented. In some examples, the data memory 108 may include a stimuli database that may store a set of stimuli to be presented, such as images, sounds, and / or strings. In some examples, the data memory 108 may include a stimuli database that may store a set of identifiers, such as numbers, characters, or a combination thereof, that may be associated with the set of stimuli that may be included in the data memory 140 of the digital device 130. In some examples, the data may include a level of information associated with each stimulus. In some examples, the data may include information about a difficulty level of a task including the plurality of stimuli or a speed of presenting the next stimulus associated with a memory state, a threshold of memory state, or a range of memory state. In some examples, the data may include sets of response and response time for providing the corresponding response of each subject collected by the digital device 130. Each set of response and response time is further stored in association with a presented stimulus or an identifier of the presented stimulus that caused the response. The data may include at least one mathematical model. The data may include extracted signature from the responses using at least one mathematical model, from the responses related to memory processes. The data may include acomputational model for estimating a performance metric. The data may include a performance metric for each particular stimulus estimated using the computational model based on the extracted signature. In some examples, the performance metric may relate to SoF. The data may include a computational model for estimating a memory metric. A memory metric generally refers to a quantitative measure of how quickly a piece of information would be forgotten by a patient. In some examples, the memory metric of each subject may be estimated based on the performance metrics of the stimuli. While a single memory device 106 is shown in FIG. 1, it is to be understood that any number of memory devices may be used, and the executable instructions and / or data may be distributed across multiple memories accessible to the processor 104.

[0052] Examples of computing devices described herein may include additional components. For example, the computing device 102 may include or be coupled to output devices. In some examples, the output devices may be one or more display(s), such as a display 126 of FIG. 1, and / or speakers. While FIG. 1 shows the display 126 integrated into the computing device 102, the display 126 or any output devices may be external devices coupled to the computing device 102. For example, the computing device 102 may include or be coupled to input devices. In some examples, the input device(s) may include keys, buttons, keyboards, mice, touchscreens, microphones, etc. The additional components in the computing device 102 may communicate with the processor 104 and / or the memory device 106 of FIG. 1 via the internal bus 124. The computing device 102 may further include communication interface 128 (e.g., cellular antenna, Wi-Fi, network interface such as the Internet, Wi-Fi, LAN, WAN, BT, NFC) of FIG. 1 that may communicate wirelessly or via wire(s) such as USB cables, ether cables, HDMI cables, or other standardized or proprietary cables. The additional components and / or the digital device 130 coupled to the computing device 102 may communicate with the computing device 102 via the communication interface 128. The communication interface 128 may handle communications between the computing device 102 and external devices, including the digital device 130.

[0053] A set of stimuli or each stimulus in a set of stimuli may be selected. In some examples, the executable instructions for determining stimulus 122 in the executable instructions for determining memory metrics 112 may cause the processor 104 of the computing device 102 to select a stimulus and / or a course of stimuli during a session. The stimuli may be, for example, questions regarding a meaning of one or more words, a prompt to identify one or more locations on a map, or a prompt for a recall of factual information. In some examples, the computing device 102 may provide the digital device 130 with the selected stimulus and / or the course of stimuli during the session in a form of images, sounds,and / or strings. In some examples, the computing device 102 may provide the digital device 130 with a corresponding identifier, such as numeric, characters, or a combination thereof, of the selected stimulus and / or a set of identifiers that represents a corresponding course of stimuli. In some examples, the set of stimuli may be stored in the data memory 140 associated with a set of corresponding identifiers. The executable instructions for presenting stimuli 144 may cause the processor 136 of the digital device 130 to prepare a next stimulus or a course of stimuli based on the received identifier or set of identifiers. In some examples, executable instructions for determining stimulus 122 may be included in the executable instructions for presenting stimuli 144 stored in the digital device 130. The executable instructions for presenting stimuli 144 may cause the processor 136 of the digital device 130 to select a next stimulus or a course of stimuli independently. Thus, a stimulus or a course of stimuli may be selected by either the computing devices 102 or the digital device 130.

[0054] In some examples, a session may include an open-ended, continuous series of behavioral responses by a subject to a series of corresponding experimental stimuli presented over a software interface (e.g., displayed on a digital device, sound produced from the digital device, etc.). During a session, each selected stimulus may be presented once or multiple times. In some examples, the executable instructions for presenting stimuli 144 performed by the processor 136 may cause the stimuli presentation circuitry, such as the display 132, to present the selected stimulus. In some examples, the executable instructions for collecting responses 146 stored in the digital device 130 may cause the processor 136 of the digital device 130 to collect each corresponding response to each stimulus. In some examples, upon presenting a stimulus, the digital device 130 may wait for one or more inputs, such as keystrokes, mouse clicks, spoken responses, and / or touchscreen gestures of the subject 150 from data acquisition circuitry, such the input interface 134. Concurrently, the executable instructions for collecting responses 146 may cause the processor 136 to run a timer to measure a response time (e.g. the time passed from the presentation of the stimulus to the response). Upon detecting the one or more inputs at the input interface 134, the executable instructions for collecting responses 146 may cause the processor 136 of the digital device 130 to record the corresponding time measured as the response time. Thus, behavioral responses including a response to the stimulus selected by the subject and its associated response time may be collected. Sets of responses and response times by each subject responsive to the presented set of stimuli may be collected for each subject throughout a session. In some examples, the digital device 130 may transmit the collected sets of response and response time to the computing device 102. The digital device 130 may provide the responses to the communication interface 148, and the communication interface 148 maytransmit the sets of the response and response time to the communication interface 128 either wired connection or wireless connection described herein. The computing device 102 may add the received sets of response and response time to the temporary record of all sets stored in the data memory 108 through the session in association with the presented stimuli.

[0055] A signature related to one or more memory processes may be extracted from the responses. In some examples, the executable instructions for extracting signature from responses 116 may cause the processor 104 to extract signature from the behavioral responses related to memory processes. In some examples, the processor 104 may separate non-memory components and memory retrieval components of a particular response. In some examples, the subject 150 may provide a behavioral response to a stimulus with a particular amount of time (e.g., a response time). Some portion of the response time may be due to perception response time to recognize or sense (e.g., see, hear) the stimulus and motor response time to physically provide a response (e.g., locate the input device and make an action with respect to the input device). Some portion of the response time may include time for memory retrieval components, such as time for retrieval of a behavioral response to the stimulus. The processor 104 may extract signature from the behavioral responses related to memory processes, such as the time for retrieval of an answer, using at least one mathematical model stored in the data memory 108. In some examples, the at least one mathematical model may be trained to separate aspects of the behavioral response attributable to a memory retrieval phase from aspects attributable to other phases, such as the perception response time and motor response time. That mathematical model could be trained on training data including earlier collected likelihood distributions of relative phase duration. In some examples, the mathematical model may include a Linear Ballistic Accumulator (LBA). The LBA may be fit to the collected behavioral responses to output the signature. In some examples, the mathematical model may include Drift-Diffusion models and / or any derived variants. In some examples, the mathematical model may apply Bayesian estimation techniques to balance prior knowledge of likely phase duration against currently observed durations, MLE-based estimation of stimulus-specific parameters, etc. Thus, the processor 104 may separate times for non-memory processes of the subject 150 from a time for memory retrieval in the response time. In some examples, the non-memory processes may include the perception and motor components, such as time for visually recognizing the stimuli prior to memory retrieval and time for physically providing the response after the memory retrieval. The processor 104 may provide a response time estimated to be due to memory retrieval upon filtering perception and motor response time, etc., with the response of the subject 150. The processor 104 may store the extracted signature from the responses in the data memory 108. While the example ofextraction of signature described herein may be performed by the processor 104, the execution of signature extraction may be performed by another processor and may not be limited to the processor 104 of the computing device 102. In some examples, the executable instructions for extracting signature from responses 116 and the mathematical model may be stored in the memory device 138 of the digital device 130 that cause the processor 136 to perform the signature extraction. These pre-memory retrieval and post-memory retrieval times may help effective comparison of memory retrieval performance across subjects, because subjectdependent perception and motor response times, irrelevant to memory retrieval, may often be salient and obscure analysis of memory retrieval performance.

[0056] A performance metric for each stimulus may be calculated and / or estimated using a computational model based on the extracted signature during the session. In some examples, the performance metric may relate to SoF. In some examples, the executable instructions for estimating a performance metric related to speed of forgetting 118 may cause the processor 104 to estimate a performance metric for each stimulus based on the extracted signature using at least one computational model stored in the data memory 108. The processor 104 may store the performance metric in the data memory 108. While the performance metric estimation described herein may be performed by the processor 104, the execution of performance metric estimation may be performed by another processor, and may not be limited to the processor 104 of the computing device 102. In some examples, the executable instructions for estimating a performance metric related to speed of forgetting 118 and the computational model may be stored in the memory device 138 of the digital device 130 that cause the processor 136 to perform the performance metric estimation.

[0057] A memory metric of the subject 150 may be estimated based on performance metrics obtained from the subject 150, using a computational model. In some examples, the executable instructions for estimating a memory metric based on performance metrics 120 may cause the processor 104 to estimate a memory metric for each subject 150 based on the performance metrics using at least one computational model stored in the data memory 108. The processor 104 may store the memory metric in the data memory 108. While the memory metric estimation described herein may be performed by the processor 104, the execution of memory metric estimation may be performed by another processor and may not be limited to the processor 104 of the computing device 102. In some examples, the executable instructions for estimating a memory metric based on performance metrics 120 and the computational model may be stored in the memory device 138 of the digital device 130 that cause the processor 136 to perform the memory metrics estimation.

[0058] In some examples, based on the memory metric, a next stimulus of a set of stimuli to be presented may be adjusted. The executable instructions for determining memory metrics 112 may cause the processor 104 to adjust either selection a next stimulus or to a manner of presentation of a stimulus. In some examples, such adjustment may be performed to increase an information gain in updating the memory metric associated with memory capacity of the subject 150. In some examples, such adjustment may include adjustment of a difficulty level of a task, including a set of stimuli, that may result in selecting a different set of stimuli to be presented. In some examples, such adjustment may include adjusting a speed of presenting the next stimulus which may have been selected.

[0059] The memory metric obtained with diagnostic-related information may be provided to a medical practitioner and / or a subject. In some examples, the executable instructions for determining memory metrics 112 may cause the display 126 to present the memory metric and the diagnostic-related information. In some examples, the diagnostic-related information may be obtained by the processor 104. In some examples, the diagnostic-related information may be provided by one or more medical practitioners prior to determining the memory metric or concurrently by examining the memory metric and other information. In some examples, the digital device 130 may cause the display 132 to present the memory metric and the diagnostic-related information.

[0060] It should be understood that this and other arrangements and elements (e.g., machines, interfaces, function, orders, and groupings of functions, etc.) can be used in addition to or instead of those shown, and some elements may be omitted altogether. Various functions described herein as being performed by one or more components may be carried out by firmware, hardware, and / or software.

[0061] FIG. 2 is a block diagram of an example system 200 for determining memory metrics in accordance with examples described herein. In some examples, determining memory metrics may be performed as a portion of quantifying a patient-specific long-term memory function. Example operations of the system 200 for determining memory metrics to support the functionality and relevant design decisions are described herein. Examples described herein are related to the method used to estimate an individual’s memory metric automatically, rapidly, efficiently, and without human supervision.

[0062] The example system 200 may include a smartphone 202 and a computing device 212. The components of FIG. 2 are exemplary. Additional, fewer, and / or different components may be used in other examples. In some examples, the smartphone 202 may be wholly and / or partially implemented using the digital device 130 of FIG. 1. In some examples, thecomputing device 212 may be wholly and / or partially implemented using the computing device 102 of FIG. 1. The computing device 212 may perform executable instructions 214. In some examples, the executable instructions 214 may be wholly and / or partially implemented as the executable instructions for determining memory metrics 112 of FIG. 1. In some examples, the smartphone 202 and the computing device 212 may be different devices. In some examples, the smartphone 202 and the computing device 212 may be integrated as one device.

[0063] In some examples, the smartphone 202 may include an input / output interface 204 which may present a stimulus 206 and receive a response 208. In some examples, the input / output interface 204 may be implemented wholly and / or partially implemented using the display 132 and the input interface 134 of FIG. 1. In some examples, the input / output interface 204 may include separate components, such as a display, a touch pad, and / or a speaker, or an audio / visual interface coupled to external video / audio interface. In some examples, the input / output interface 204 may include a touchscreen for receiving a response 208. The response 208 may include, for example, touchscreen gestures. In some examples, the input / output interface 204 may be implemented using another input interface 134 and the response 208 may include keystrokes, mouse clicks, spoken responses, etc.

[0064] In some examples, the executable instructions for presenting stimuli 144 of FIG. 1 may be performed by the smartphone 202. The executable instructions for presenting stimuli 144 may cause the stimuli presentation circuitry, such as the input / output interface 204, to present the selected stimulus 206. Examples of a stimulus may include an item that a subject is asked to remember during a session. A set of stimuli may include items (e.g., words, images, objects, sounds, etc.) that are not overly familiar to the subject. In some examples, the set of stimuli may be visual, textual, verbal, or auditory. In some examples, the stimulus may be presented through the input / output interface 204. The input / output interface 204 may include, for example, a screen display for presenting the stimulus including visual or textual information or a sound output speaker for presenting the stimulus including verbal or auditory information.

[0065] The input / output interface 204 may collect the behavioral response 208 following presentation of the stimulus. In some examples, the executable instructions for collecting responses 146 of FIG. 1 may be performed by the smartphone 202. The executable instructions for collecting responses 146 may cause data acquisition circuitry, such as the input / output interface 204, to collect each corresponding behavioral response 208 to each stimulus 206 provided by a subject. In some examples, upon presenting the stimulus 206, the smartphone 202 may wait for one or more inputs, such as keystrokes, mouse clicks, spoken responses,touchscreen gestures, from the input / output interface 204. For example, the response 208 may be recorded as one or more keystrokes from a keyboard, a gesture performed by pressing a finger on a touchscreen, a click or a movement on a pointing or tracking device, and / or a vocal response to a microphone. Concurrently, the executable instructions for collecting responses 146 may cause the smartphone 202 to run a timer to measure a corresponding time elapsed for such inputs (e.g., response time).

[0066] The response 208 and the corresponding response time may be combined as a set 210 of response 208 and the corresponding response time and provided to the computing device 212. Sets 210 of the response 208 and corresponding response time provided by each subject responsive to the presented stimuli may be collected for each subject throughout a session by the smartphone 202 and transmitted to the computing device 212. The smartphone 202 may provide the computing device 212 the responses 208 using either wired connection (e.g., USB cables, ether cables using TCP / IP protocol, HDMI cables, or other standardized or proprietary cables) or wireless connection (e.g., cellular antenna, Wi-Fi, network interface such as the Internet, Wi-Fi, LAN, WAN, BT, NFC) described herein. In some examples, the smartphone 202 and computing device 212 may be integrated and transmission may be performed using an internal bus. In some examples, the smartphone 202 and computing device 212 may communicate the responses 208 via a memory storage device, either within one of the smartphone 202 and computing device 212 or located outside of the smartphone 202 and computing device 212.

[0067] The computing device 212 may perform executable instructions 214. The executable instructions 214 may include, for example, the executable instructions for determining memory metrics 112. The executable instructions for determining memory metrics 112 may determine memory metrics 224 using one or more computational models 222.

[0068] The one or more computational models 222 will be explained. In some examples, the one or more computational models 222 may assume that a memory includes individual traces created every time the same information is encountered. Each trace may decay according to the power law of forgetting. FIG. 3 shows an equation 300 representing log odds of retrieving a memory in accordance with examples described herein. The log odd of retrieving a memory m at time t is proportional to its activation A(m, f). The equation 300 may compute the “activation” of a memory m at time / , denoted as A(m, f). The activation may change as a function of both time and the responses given by a subject to each presentation of a stimulus.

[0069] FIG. 4 shows an equation 400 representing a trace-specific decay rate in accordance with examples described herein. In the equation 400, ti is the time associated with thepresentation of the stimulus, a is stimulus- and subject-specific SoF, and c is a fixed parameter. The stimulus- and subject-specific SoF a captures the subject’s ability to memorize said stimulus. The SoF value that is characteristic of a subject might be estimated by presenting multiple stimuli. The fixed parameter c may have been predetermined by analyzing large amounts of data from published or unpublished studies to represent average human behavior. In some examples, c is 0.25. In some examples, the fixed parameters c may have a different value. Where f is the creation time of the i-th trace, and d(i) is a characteristic power decay rate of the i-th trace. This trace-specific decay rate may depend on the residual activation of the memory at the time of creation of the trace. Because the decay rate of each trace in the equation 400 may depend on the memory’s activation, the equation 400 may provide an explanation for a spacing effect, where traces closer in time may have higher decay rates due to the greater activation A (m, / ) of the memory at time t(i).

[0070] As shown in the equation 400, the one or more computational models may depend on the stimulus-specific SoF “a.” The stimulus-specific SoF “a” may represent the relationship between the history of a memory and the likelihood of being able to retrieve the memory in the future.

[0071] The odds of being able to recall a memory at a later time may depend solely on the rate at which the memory is forgotten. This may also suggest that by analyzing the history of a memory and the number of times it has been assessed, the rate at which that memory is forgotten may be determined. FIG. 5 shows an equation 500 representing a probability of a correct response to a stimulus related to the activation in accordance with examples described herein. In the equation 500, r is an individual parameter, known as a speed-accuracy tradeoff, that may determine the degree of certainty to be achieved in responses expected by a subject. The fixed parameter r may have been predetermined by analyzing large amounts of data from published or unpublished studies and finding to represent average human behavior. In some examples, r is 0.80. In some examples, the fixed parameter r may have different values.

[0072] FIG. 6 shows an equation 600 representing a response time associated with a correct response to a stimulus related to the activation in accordance with examples described herein. In the equation 600, time to is a combination of time associated with perceiving a stimulus and general motor agility of a subject. The parameter to may be estimated for a single session or a portion of the session of a single patient. The time associated with perceiving the stimulus may be affected by, for example, dyslexia when the stimulus is among a set of textual stimuli. When a subject is slower with a lower general motor agility, the subject may have higher values of to. For example, elderly individuals may have less motor agility, and therefore higher values of to, than younger individuals. The fixed parameter F might scale the time it takes togive a response based on the retrieved memory. For example, elderly individuals might take longer times to access memories, or be more cautious in responding, even when the activation of a memory is the same. The value of F may have been predetermined by analyzing large amounts of data from published studies and finding to represent average human behavior. In some examples, F is 1.0. In some examples, the fixed parameter F may have different values. In some examples, the parameter F might be different for every subject.

[0073] The average SoF across a set of stimuli may be a characteristic of a subject representing a memory metric of their memory function. In some examples, the memory metric is highly stable (r > 0.7) across times and materials. Furthermore, in some examples, using neuroimaging methods, the SoF may represent individual differences in long-term memory function. In some examples, the SoF may correlate with, and can be decoded from, spontaneous brain activity at rest from a subject. As discussed earlier, the model contains several parameters. The parameters c, F, and r may have been predetermined by analyzing large amounts of data from published or unpublished studies and finding the values that, when included in equations 300-600, represent average human behavior.

[0074] Returning to FIG. 2, the computing device 212 may add the collected sets 210 of response 208 and response time in association with the presented stimuli 206 to the temporary record of sets of response 208 and response time upon receiving responses 216. In some examples, if the record of sets 210 of response 208 and response time for that particular stimulus does not contain any data, an initial default value of the SoF a in the equation 400 may be used. The initial value might or might not be informed by other information available about the subject (e.g., age, sex, level of education, previous clinical history). If the record of sets 210 of response 208 and response time for that particular stimulus includes any data, then an existing estimate of the SoF a of the stimulus may have been computed. Based on the response 208, the SoF a for the stimulus is then updated, either upwards or downwards.

[0075] In some examples, during the updating of the SoF, the contribution of the parameter to may be separated from the contribution of the A(m, t) to the response times. The computing device 212 may extract a signature related to one or more memory processes from the set 210 of response 208 and response time. In some examples, non-memory components and memory retrieval components of a response time may be separated using at least one mathematical model stored, such as a mathematical model 218. In some examples, the at least one mathematical model 218 may be trained to separate aspects of the behavioral response attributable to a memory retrieval phase from aspects attributable to other phases, such as the perception response time and motor response time. In some examples, the mathematical model 218 may be based on the equation 600. The mathematical model could be trained on trainingdata including earlier collected likelihood distributions of relative phase duration. In some examples, the mathematical model may include an LBA. The LBA may be fit to the collected behavioral responses to output the signature. In some examples, the mathematical model may include Drift-Diffusion models and / or any derived variants. In some examples, the mathematical model may apply Bayesian estimation techniques to balance prior knowledge of likely phase duration against currently observed durations, MLE-based estimation of stimulus-specific parameters, etc. The processor 104 may segment the time to of the equation 600 for non-memory processes and a time for memory retrieval in the response time in each set 210, and extracting signature 220 may be performed by separating (e.g., removing, subtracting, segmenting) the response time for non-memory processes from the response time. Thus, the time to is then separated (e.g., removed, subtracted, segmented out) from affected response times in the record of sets 210 of the responses 208 and their associated times for the set of corresponding stimuli.

[0076] While the example of extraction of signature described herein may be performed by the computing device 212, the execution of signature extraction may be performed by another processor, and may not be limited to the computing device 212. In some examples, the mathematical model 218 may be stored in the smartphone 202 to perform the signature extraction. These pre-memory retrieval and post-memory retrieval times may help effective comparison of memory retrieval performance across subjects, because subject-dependent perception and motor response times, irrelevant to memory retrieval, may often be salient and obscure analysis of memory retrieval performance.

[0077] After the response time for any stimulus is adjusted, a re-evaluation of the patient’s memory metric using computational models 222 may be performed through numerical simulations of the sets of responses and their associated time and the corresponding stimuli through the session. The updating of the SoF a for the stimulus may be performed through numerical simulations of the expected behavioral responses 208 using the computational models 222. A performance metric for each stimulus may be estimated using a computational model in the computational models 222 based on the extracted signature during the session. In some examples, the performance metric may relate to SoF. In some examples, the stimulusspecific SoF a may be estimated for each stimulus presented to a patient within a single session based on the equation 400. In some examples, the executable instructions for estimating a performance metric related to speed of forgetting 118 may cause the computing device 212 to estimate a performance metric for each stimulus based on the extracted signature using at least one computational model of the computational models 222. The computing device 212 may store the performance metric. While the performance metric estimationdescribed herein may be performed by the computing device 212, the execution of performance metric estimation may be performed by another processor, and may not be limited to the computing device 212. In some examples, the executable instructions for estimating a performance metric related to speed of forgetting 118 and the computational model may be stored in the smartphone 202 to cause the smartphone 202 perform the performance metric estimation.

[0078] A memory metric of each subject may be estimated, using a computational model, based on performance metrics obtained from the subject in the computational models 222. In some examples, the executable instructions for estimating a memory metric based on performance metrics 120 may cause the computing device 212 to estimate a memory metric 224 for each subject based on the performance metrics using at least one computational model in the computational models 222. The computing device 212 may store the memory metric 224. In some examples, one or more SoF a may be aggregated together into a single, representative memory metric A(m, t) of a subject based on the equation 300. In some examples, the distributions of correct and incorrect responses 208 and their associated response times are modeled as an evidence accumulation process whose accumulation rate is proportional to the memory’s activation A(m, t) based on the equation 300. While the memory metric estimation described herein may be performed by the computing device 212, the execution of memory metric estimation may be performed by another processor, and may not be limited to the computing device 212. In some examples, the executable instructions for estimating a memory metric based on performance metrics 120 and the computational model may be stored the smartphone 202 may cause the smartphone 202 to perform the memory metrics estimation.

[0079] In some examples, based on the memory metric, a next stimulus of a set of stimuli to be presented may be adjusted. The executable instructions 214 may cause the computing device 212 to adjust either selection a next stimulus or a manner of presentation of a stimulus. In some examples, such adjustment may be performed to increase an information gain in updating the memory metric associated with memory capacity of the subject 150. In some examples, such adjustment may include adjustment of a difficulty level of a task, including a set of stimuli, that may result in selecting a different set of stimuli to be presented. In some examples, such adjustment may include adjusting the interval at which to present the next stimulus which may have been selected. A next stimulus 228 to be presented may be one of the previously used stimuli or a new stimulus that has never been previously used in the session. In some example, a next stimulus 228 may be selected from a stimuli database 226. For example, if the stimulus is a new stimulus that has not been previously used, the newstimulus may be retrieved from the stimuli database 226. In some examples, the stimulus may be selected based on its utility in reducing the uncertainty around the value of the memory metric. For example, if the SoF a for one of the previously used stimuli is associated with unusually high or low a values, the method might present the stimulus again to collect more responses that may aid in estimating a correct value by detecting abnormal a value. Alternatively, if all the previously used stimuli are associated with consistent values, the executable instructions 214 may select a new stimulus. If no new stimulus (e.g., has not been presented) exists in the stimuli database 226 or time elapsed since the beginning of the session has exceeded a predefined time limit, the session may terminate.

[0080] Once the session ends, a final estimate for a memory metric 224 of the subject may be saved and provided. The memory metric 224 obtained with diagnostic-related information may be provided to a medical practitioner and / or a subject. In some examples, a display of the computer computing device 212 may present the memory metric and the diagnostic-related information. In some examples, the input / output interface 204 (e.g., a touchscreen) of the smartphone 202 may present the memory metric and the diagnostic-related information.

[0081] In some examples, the diagnostic-related information may be provided by one or more medical practitioners prior to determining the memory metric or concurrently by examining the memory metric and other information. The returned memory metric 224 may be used by a physician to make a number of health-related decisions about the patient. For example, a physician may compare a patient’s memory metric against a threshold that is agreed upon as being indicative of AD or other (sub-)clinical conditions and use the memory metric to reach a diagnosis. Alternatively, a physician might compare a patient’s memory metric to other patients of similar age and condition, and, if the memory metric appears suspiciously higher than the other patients’, may decide to recommend further neurological examinations (e.g., screening / scanning using computed axial tomography or magnetic resonance imaging). Alternatively, a physician may use a patient’s memory metric to assess the impact of a procedure (e.g., brain surgery or medication) on the cognitive function of a patient. The physician may collect memory metrics of a patient before and after the procedure and compare the two. If the procedure was intended to alleviate conditions that had affected a patient’s memory, the physician may expect the memory metric to decrease after surgery. Alternatively, if the procedure had accidentally touched upon the patient’s memory circuits, the physician may expect the patient’s memory metric to increase. Alternatively, a physician might use the memory metric to evaluate whether the impact of a clinical event (e.g., a mild traumatic brain injury) or a cognitive state (e.g., a depressive episode) has subsided. Evenwhen no prior metric of a patient is available, observing a relatively high SoF followed by a decrease and then a new plateau at a lower level might indicate recovery.

[0082] Examples described herein include systems and methods that may quantitatively assess a patient’s long-term memory function. Unlike traditional questionnaires, examples described herein may be automatic, without human supervision, language or cultural standardization, and less prone to practice effects. Instead of using an arbitrary measure of memory performance and comparing it to a standardized sample, examples described herein quantify the speed at which processes of forgetting occur in the brain. Example methods may quantify the SoF by fitting parameters of a mathematical model of passive forgetting to data collected from a subject. The data is collected from a digital device that interacts with a subject for a brief amount of time and provides responses to stimuli. The one or more models may be fitted online and interactively, refining its parameters as responses come in. In some examples, example methods use the current estimates of a patient’s memory function to select the most appropriate stimulus to present next.

[0083] FIG. 7 shows a diagram 702 of predicted forgetting associated with a specific value of the memory metric in accordance with examples described herein. FIG. 7 may be obtained through the procedure described with the system 200 of FIG. 2, and may utilize the system shown in FIG. 1 and / or FIG. 2. In the diagram 702, four curves are traced, each curve corresponding to the predicted forgetting associated with a specific value of the memory metric. Any point on a curve represents the probability of forgetting a fact presented after a certain amount of time elapsed. The four points represent times at which a fact presented is forgotten with 95% probability for four different values of the memory metric. The four values correspond to the typical memory metrics observed in (a) college undergraduates (metric = 0.29); (b) middle-aged individuals (metric = 0.33); (c) individuals affected by MCI (metric = 0.42); and (d) individuals affected by AD (metric = 0.53).

[0084] Unlike other tests, examples of a memory metric described herein are directly interpretable in terms of the changes in the probability of forgetting over time. Different memory metrics correspond to different curves on a plot that tracks the probability of forgetting against time. These memory metrics can be plotted against any reference data (for example, data from a cohort of individuals with the same age, educational background, and gender as the patient) or historical data from the same patient. Memory metrics are stable across repeated tests and materials. Unlike other measures of memory function, examples of the memory metric described herein are directly interpretable and do not require standardization and normative metrics. As implied by the equations 300-500, the memorymetric is indicative of how quickly the probability of remembering declines over time for a particular patient.EXPERIMENTS

[0085] An example method for determining a memory metric was implemented and tested on 16 individuals. FIGS. 8-16 may be obtained through the procedure described with the system 200 of FIG. 2, and may utilize the system shown in FIG. 1 and / or FIG. 2. Results of the test demonstrated that, on average, the correlation between different values of the memory metric, obtained from two different sessions with different materials and separated by an interval between one and 26 weeks, is 0.71.

[0086] Experimental Scenarios

[0087] Based on the behavioral and imaging findings, The average SoF successfully represents the different biological processes of passive forgetting at a computational level. These processes include loss of context clues, retrieval interference from other similar memories, and “natural” biological decay. Some of these processes are accelerated in aging and abnormally elevated in amnestic dementias, such as AD. Thus, the model described herein may distinguish between abnormal memory impairments and normal aging controls. Accordingly, this model could become a useful tool in the clinical assessment of memory function. Furthermore, the model-based assessment may provide a new, highly detailed view of memory decay trajectories in normal and abnormal aging, and of the effects of interventions.

[0088] A longitudinal study of healthy elderly adults and elderly individuals with MCI was conducted. Individuals with MCI, rather than dementia, were selected as participants for having sufficient cognitive abilities to perform tasks in the experiment. Thus, the models’ ability to detect earlier, more subtle differences in memory function may be tested as such differences may often be a precursor to AD and other forms of dementia. This cohort of individuals was followed for over six months, during which they performed weekly online model-based assessments to characterize their SoF. The scenarios include: (1) individuals with MCI may exhibit greater SoF values than healthy controls; (2) SoF values would be reliable across repeated assessments; (3) SoF values may have clinical validity (e.g., possibility to identify differences in abnormal memory function from an individual SoF); and (4) SoF may increase over a period of months, capturing a trajectory of abnormal and healthy aging.

[0089] Materials and Methods

[0090] Sixteen participants were recruited on a rolling basis from the local NIH-designated Alzheimer’s Disease Research Center. Inclusion criteria for the study include: (1) age between 55 and 85 years; (2) fluency in English; and (3) no major medical or psychiatric conditions that may affect cognitive performance. Participants were classified into two groups: healthy cognition (a number of subjects 7; 3 females aged 58-71, 4 males aged 57-71) and those with MCI (a number of MCI subjects 9; 2 females aged 63, 7 males aged 67-78).

[0091] MCI can be defined as a decline in cognitive abilities that is greater than what is typical for a person’s age and educational background, but does not meet the criteria for a diagnosis of dementia. MCI was diagnosed using a combination of methods including clinical evaluation, cognitive testing, and medical history. The clinical evaluation was conducted by a geriatric psychiatrist or a neurologist, who assessed the participant’s cognitive and functional abilities using standardized tools. Cognitive testing was performed using a battery of neuropsychological tests that measured various cognitive domains such as memory, attention, and executive function. Medical history was obtained through a structured interview and review of medical records. Participants were classified as having MCI if they had a Clinical Dementia Rating scale <= 0.5. Additionally, individuals with subjective reports of decline by self and / or informant in conjunction with objective cognitive deficits were also included in the MCI group. Healthy controls were screened for cognitive impairment using the same methods as for MCI participants. They were classified as healthy controls if they scored within normal limits on cognitive tests and had no history of cognitive decline or functional impairment. All participants provided informed consent and were compensated for their participation in the online memory game portion of the study.

[0092] Adaptive Memory Assessment

[0093] Weekly at-home assessments were completed with an online adaptive fact learning system (AFLS). This system continuously estimates individualized SoF values in real time as the participant works through the lesson. The software was designed so that participants could perform the task from home using any mobile device. The AFLS works by presenting new study pairs (e.g., “France / Paris”) and scheduling repeated tests (e.g., “France / ?”) at strategic points based on the online estimates of a user’s SoF. FIG. 8 shows examples of a software interface in accordance with examples described herein. In some examples, the software interface may be a screen 804 on a smartphone 802. The components of FIG. 8 are exemplary. Additional, fewer, and / or different components may be used in other examples. In some examples, the smartphone 802 may be wholly and / or partially implemented using the digital device 130 of FIG. 1 and / or the smartphone 202 of FIG. 2. The smartphone 802 may present a stimulus on its screen 804, showing an image of pasta and a text string “Conchiglie,” andthen present a test showing multiple choices in text, including the text string “Conchiglie” for a subject to select. After a while, a next stimulus may be presented on the screen 804, showing an image of another type of pasta and a text string “Fusilli .”

[0094] Study Materials

[0095] Thirty-two lessons were prepared in advance, spanning different topics (such as European capitals, Swahili words, Asian flags, bird species, types of pasta, flower species). The materials were vetted prior to the experiment to make sure they were comparable in terms of familiarity and difficulty. For each lesson, 15 different pairs were created, each of which associated an object with an English noun. In half of the pairs, the object was presented as an image (e.g., a picture of a sterling with the name “Sterling” for a lesson of birds), and in the other half, the object was a word (e.g., “France” / “Paris” for European capitals). This was done to investigate possible differences due to the encoding modality (e.g., verbal vs. visual objects). The number of terms reached in each lesson depended on the response times and errors of the individual.

[0096] Data Processing

[0097] The repetition, activation, and SoF values for each term were calculated using functions from the software package. The average SoF values for each lesson and individual were identified by a value of each pair at the very last repetition of that term. The data was then filtered to only contain the first full session of a topic that is over six minutes. This length of the session was designed to eliminate any superfluous sessions when some participants desired to complete the task more than once). The data was also organized by the week the lesson was completed to view temporal trends.

[0098] Analysis

[0099] FIG. 9 includes diagrams 902 and 904 showing a distribution of SoF values across all lessons and patients in accordance with examples described herein. The diagram 902 shows individual SoF on a horizontal axis and a number of observations on a vertical axis for stimuli across topics and materials. For all the stimuli, individual SoFs varied between 0.29 and 0.58 and were normally distributed. The diagram 904 shows ranges and means of SoF across lessons ranged from 0.29 to 0.55 with a mean of 0.4.

[0100] The test-retest reliability of the SoF across materials was assessed using pairwise Pearson correlations between every pair of lessons. FIG. 10 is a diagram 1002 showing correlations between memory metrics measured across lessons of different materials. The SoFs across lessons of different materials were found to have a high average correlation coefficient of r = 0.70.

[0101] Global differences in SoF for healthy controls and individuals diagnosed with MCI as per the gold standard clinical assessment were compared. FIG. 11 includes diagrams 1102 and 1104 showing a distribution of memory metrics across patients affected by MCI and age- matched healthy controls in accordance with examples described herein. The diagram 1102 shows individual SoF on a horizontal axis and a number of observations on a vertical axis for MCI and healthy controls subject groups. On average, healthy controls had a mean of SoF that was 0.39, whereas MCI had a mean of SoF that was 0.42. The differences in means and ranges between the subject groups are observed consistently through topics as shown in the diagram 1104. The difference between groups was compared using a mixed-effects linear model that included the specific weekly topic as a random effect to account for differences in familiarity. The linear model confirmed the existence of a large main effect of group (P = 0.04, t = 9.78, p<0.0001).

[0102] Diagnostic Validity of the SoF

[0103] To analyze the parameter’s diagnostic potential, its classification accuracy was plotted with a ROC curve. The ROC curve assesses the sensitivity (true positive rate) and specificity (true negative rate) of a classifier for varying thresholds of the SoFs. The overall accuracy of the classifier is then measured as an area under the curve (AUC) of the sensitivity and specificity obtained for different thresholds.

[0104] FIG. 12A shows ROC classification performances for SoF and for response accuracy from a single session in accordance with examples described herein. The ROC curves in FIG. 12A for a single eight-minute session of data including probability of correctly identifying group members (MCI or controls) by the SoF of a single test were examined. The model proved to be highly diagnostic, with an AUC = 0.786 (equivalent to a classification accuracy of 78.6%).

[0105] Then, the ROC curves for a classifier built on the average SoF of an individual computed across all sessions were examined. FIG. 12B shows average ROC classification performances for SoF and for response accuracy across sessions in accordance with examples described herein. As expected, the classifier showed an improved accuracy of 83.6% in FIG. 12B. Because of the high test-retest reliability of the SoF, classification using a single session is almost as accurate as when averaging over 30+ sessions.

[0106] Whether the SoF provided additional validity over traditional behavioral measures, such as response accuracy, was investigated. Accuracy and response times are collected from the AFLS, which already uses the SoF to determine the best moment at which a study pair is presented. Thus, the validity of these traditional behavioral measures is, in fact, inflatedbecause they benefit indirectly from the study items being driven by the SoF. Despite this, a classifier based on response accuracy alone had a lower AUC than a classifier built on SoF, both for single sessions in FIG. 12A and for aggregated data in FIG. 12B. Specifically, in both cases an SoF -based classifier showed greater sensitivity than an accuracy-based one even at high levels of specificity, as shown in lower left corners of FIGS. 12A and 12B. From these results, SoF appears to demonstrate greater ability to detect mild memory impairments while avoiding false positives.

[0107] Finally, a logistic curve was generated to model the probability of the binary diagnosis outcomes. FIG. 13 shows probability of MCI by different levels of memory metrics in accordance with examples described herein. To generate the model, individual SoF values for each lesson were binned in increments of 0.01, and the probability of MCI diagnosis was computed as the proportion of individuals with an MCI diagnosis for each bin. The logistic model demonstrated strong fit to the data (Cragg and Uhl er’ s pseudo R2 = 0.26). The inflection point of the curve (e.g., the point at which the probability of an MCI diagnosis is > 50%) was found to be at an SoF value a = 0.40. Additionally, the curve revealed that the probability of an MCI diagnosis occurring increased steadily as the a value increased, approaching a maximum probability of 1.0 at an SoF value a = 0.52. Overall, the logistic curve provided a clear visualization of the relationship between the predictor SoF and a diagnosis of MCI.

[0108] Memory Assessment Temporal Trends

[0109] FIG. 14 shows age-related changes in memory metric over time in MCI patients and healthy controls in accordance with examples described herein. As memory function worsens in MCI patients, speeds of forgetting should steadily increase over time. While participants were only halfway through the year-long experiment, the subtle changes in the longitudinal trajectory of MCI patients can already be seen. The effect of time was captured using a mixed- effects linear model that included the week number as the main factor and the weekly topic as a random effect (to account for differences in difficulty across topics). In other words, forgetting grew significantly in samples over time, by approximately 0.15% per week. However, no significant interaction between time and group existed. The analysis uncovered a significant effect of the week (P = 0.0005, t = 2.06, p = 0.04). Further analysis showed that this effect was driven by the MCI group alone; separate linear models uncovered a significant effect of the week for the MCI (P = 0.0006, t = 2.02, p = 0.04) but not for healthy controls (P = 0.0003, t = 1.13, p > 0.26). Thus, although the SoF value increased almost twice as fast than in controls, a significant difference in the rate of growth could not yet be detected within limited samples and time window.

[0110] Differences Between Materials[OHl] FIG. 15 shows a distribution of memory metric values across different groups and types of stimuli in accordance with examples described herein. As noted above, this longitudinal study also offered the opportunity to examine the effects of different types of materials (e.g., verbal vs. visual) on memory function. To assess if the type of stimulus material presented affected lesson difficulty, SoFs for verbal stimuli and visual stimuli lessons were compared. Healthy controls had similar SoF averages (verbal a = 0.390; visual a = 0.383); on the other hand, MCIs had a greater SoF value for verbal stimuli (verbal a = 0.430; visual a = 0.416). Upon further investigation, the main stimulus features contributing to differences in verbal and visual materials for MCIs appeared to be non-English language and numeracy, both of which were primarily in verbal materials. However, there may be another interpretation that the visual stimuli might have provided more features to aid in memorization.

[0112] MCI Subtype

[0113] FIG. 16 shows a detailed view of memory metrics across materials and patient subgroups in accordance with examples described herein. To explore the model’s ability to parse out the nuances in cognitive deficits, the patients were categorized based on MCI group subtypes. Using SoF, MCI subtypes, including amnestic single and multiple domain (aMCI S, aMCI M) and nonamnestic MCI (naMCI), may be accurately distinguished. The aMCI subtype is characterized by a specific memory impairment, while the naMCI subtype is characterized by a more general cognitive decline. The results in FIG. 16 revealed that the cognitive profile of the naMCI participant more closely resembled that of the healthy control group. This observation is in line with the fact that naMCI is characterized by cognitive decline in domains other than memory, such as executive function (e.g., speed of processing, problem solving, set shifting, inhibition). Therefore, it is expected that the data would be more comparable to the control group as there is no memory loss present in naMCI. However, the model showed limitations in accuracy for the single dementia patient, likely due to differences in response time and reduced facts retained, making further refinement desirable for this population.

[0114] A novel approach for tracking and diagnosing mild memory impairments using the SoF from a computational cognitive model has been described herein. While this model was originally developed for student fact learning, it has never been explicitly used in clinical populations. Here, the SoF was found to be normally distributed, with higher means and ranges in MCI, and showed high diagnostic validity with test-retest reliability. The model alsorevealed differences in MCI verbal and visual memory, MCI subtypes, and subtle declines over time. Verbal versus visual differences may pinpoint left hemisphere involvement. AD is often asymmetric, yet clinics primarily use verbal memory to diagnose MCI. This model may reduce the ascertainment bias towards right-lateralized MCI. In the example of aMCI versus naMCI, SoF proved to be a purer assessment of memory impairment by avoiding confounds with retrieval strategy and executive function - which may be a problem in clinical assessment of mild memory impairment. The ability to track memory over time as early detection of MCI is likely to be useful in therapies to delay AD and related conditions, and the brief, user- friendly online format makes passive data assessments remarkably convenient. The SoF is demonstrated as having a potential for diagnostic use with repeatability / stability to test the efficacy of interventions like neuromodulation or cognitive enhancers.

[0115] From the foregoing it will be appreciated that, although specific embodiments of the disclosure have been described herein for purposes of illustration, various modifications may be made without deviating from the spirit and scope of the disclosure.

[0116] The particulars shown herein are by way of example and for purposes of illustrative discussion of the preferred embodiments of the present disclosure.

[0117] Unless the context clearly requires otherwise, throughout the description and the claims, the words “comprise,” “comprising,” and the like are to be construed in an inclusive sense as opposed to an exclusive or exhaustive sense, that is to say, in the sense of “including, but not limited to.” Words using the singular or plural number also include the plural and singular number, respectively. Additionally, the words “herein,” “above,” and “below” and words of similar import, when used in this application, shall refer to this application as a whole and not to any particular portions of the application.

[0118] Of course, it is to be appreciated that any one of the examples, embodiments, or processes described herein may be combined with one or more other examples, embodiments, and / or processes or be separated and / or performed among separate devices or device portions in accordance with the present systems, devices, and methods.

[0119] Finally, the above discussion is intended to be merely illustrative of the present method, system, device and computer-readable medium and should not be construed as limiting the appended claims to any particular embodiment or group of embodiments. Thus, while the present method, system, device and computer-readable medium have been described in particular detail with reference to exemplary embodiments, it should also be appreciated that numerous modifications and alternative embodiments may be devised by those having ordinary skill in the art without departing from the broader and intended spirit and scope ofthe present method, system, device and computer-readable medium as set forth in the claims that follow. Accordingly, the specification and drawings are to be regarded in an illustrative manner and are not intended to limit the scope of the appended claims.

Claims

CLAIMSWhat is claimed is:

1. A method comprising: presenting a plurality of stimuli to a patient; collecting behavioral responses from the patient to the plurality of stimuli; extracting a signature related to memory processes, using at least one mathematical model, from the behavioral responses; estimating, using a first computational model, a performance metric for each particular stimulus based on the extracted signature, the performance metric relating to speed of forgetting; and estimating, using a second computational model, a memory metric of the patient based on the performance metrics of the plurality of stimuli.

2. The method of claim 1, wherein the extracting the signature comprises removing at least one portion of the behavioral responses due to non-memory processes.

3. The method of claim 2, wherein the at least one portion of the behavioral responses is related to a perception retrieval.

4. The method of claim 2, wherein the at least one portion of the behavioral responses is related to a motor response.

5. The method of claim 1, further comprising adjusting a next stimulus of the plurality of stimuli based on the memory metric.

6. The method of claim 5, wherein the adjusting the next stimulus of the plurality of stimuli is to increase an information gain in updating the memory metric associated with memory capacity of the patient.

7. The method of claim 5, wherein the adjusting the next stimulus of the plurality of stimuli comprises at least one of adjusting a difficulty level of a task including the plurality of stimuli or a speed of presenting the next stimulus.

8. The method of claim 1, further comprising displaying diagnostic-related information.

9. A system comprising: at least one processor; andone or more memory devices storing instructions that, when executed by the processor, configure the system to: present a plurality of stimuli to a patient; collect behavioral responses from the patient to the plurality of stimuli; extract a signature related to memory processes, using at least one mathematical model, from the behavioral responses; estimate, using a first computational model, a performance metric for each particular stimulus based on the extracted signature, the performance metric relating to speed of forgetting; and estimate, using a second computational model, of a memory metric of the patient based on the performance metrics of the plurality of stimuli.

10. The system of claim 9, wherein the extract the signature comprises remove at least one portion of the behavioral responses due to non-memory processes.

11. The system of claim 10, wherein the at least one portion of the behavioral responses is related to a perception retrieval.

12. The system of claim 10, wherein the at least one portion of the behavioral responses is related to a motor response.

13. The system of claim 9, wherein the instructions further configure the system to adjust a next stimulus of the plurality of stimuli based on the memory metric.

14. The system of claim 13, wherein the adjust the next stimulus of the plurality of stimuli is to increase an information gain in updating the memory metric associated with memory capacity of the patient.

15. The system of claim 13, wherein the adjust the next stimulus of the plurality of stimuli comprises at least one of adjusting a difficulty level of a task including the plurality of stimuli or a speed of presenting the next stimulus.

16. The system of claim 9, wherein the instructions further configure the system to display diagnostic-related information.

17. A non-transitory computer-readable storage medium, the computer-readable storage medium including instructions that when executed by a computer cause the computer to: present a plurality of stimuli to a patient;collect behavioral responses from the patient to the plurality of stimuli; extract a signature related to memory processes, using at least one mathematical model, from the behavioral responses; estimate, using a first computational model, a performance metric for each particular stimulus based on the extracted signature, the performance metric relating to speed of forgetting; and estimate, using a second computational model, of a memory metric of the patient based on the performance metrics of the plurality of stimuli.

18. The computer-readable storage medium of claim 17, wherein the extract the signature comprises remove at least one portion of the behavioral responses due to non-memory processes.

19. The computer-readable storage medium of claim 18, wherein the at least one portion of the behavioral responses is related to a perception retrieval.

20. The computer-readable storage medium of claim 18, wherein the at least one portion of the behavioral responses is related to a motor response.

21. The computer-readable storage medium of claim 17, wherein the instructions further configure the computer to adjust a next stimulus of the plurality of stimuli based on the memory metric.

22. The computer-readable storage medium of claim 21, wherein the adjust the next stimulus of the plurality of stimuli is to increase an information gain in updating the memory metric associated with memory capacity of the patient.

23. The computer-readable storage medium of claim 21, wherein the adjust the next stimulus of the plurality of stimuli comprises at least one of adjusting a difficulty level of a task including the plurality of stimuli or a speed of presenting the next stimulus.

24. The computer-readable storage medium of claim 17, wherein the instructions further configure the computer to display diagnostic-related information.

Images