Estimating memory impairment from a mobile multimodal digital screening

Abstract

Methods are provided for determining a memory impairment probability. The method comprises receiving one or more audio recordings of an interaction by a user with a computing device during an assessment; extracting acoustic and speech timing features from the one or more audio recordings; applying a machine learning model to the acoustic and speech timing features to generate a memory impairment probability for the user; and providing a clinical decision support recommendation for a provider based on the memory impairment probability.

Description

ESTIMATING MEMORY IMPAIRMENT FROM A MOBILE MULTIMODAL DIGITALSCREENINGCROSS REFERENCE TO RELATED APPLICATION S)

[0001] This application claims the benefit of and priority to U.S. Provisional Application No. 63 / 634045, filed April 15, 2024, which is hereby incorporated by reference in its entirety.TECHNICAL FIELD

[0002] The disclosure is generally directed to the use of mobile multimodal data to estimate cognitive impairment, and in particular, a method generating an estimate of memory impairment based on multimodal data for patients at risk of mild stages of cognitive decline.BACKGROUND OF THE DISCLOSURE

[0003] Traditional cognitive screening tools do not include detection of cognitive impairment in populations at risk of or at mild stages of cognitive decline. Patients often present with cognitive complaints long after subtle symptoms began; thus, these traditional tests are often administered after the onset of overt symptoms and often lack sensitivity. Traditional assessments do not easily fit the time constraints in primary-care settings, which are frequently at the forefront of cognitive screening, without significant workflow modifications. These assessments require administration training to administer tests reliably and require expertise to interpret the results and the significance of the findings.

[0004] As such, there is a need for a solution that can capture different domains of cognitive impairment and, particularly, be able to include assessment of the domains primarily affected by the most common neurodegenerative disorders. Currently no solution utilizes multimodal digital assessments (such as voice and speech, written metrics, etc.) to detect memory impairment, a subcomponent of general cognition.BRIEF SUMMARY OF EMBODIMENTS OF THE DISCLOSURE

[0005] In one embodiment, a method comprises receiving one or more audio recordings of an interaction by a user with a computing device during an assessment; extracting acoustic and speech timing features from the one or more audio recordings; applying a machinelearning model to the acoustic and speech timing features to generate a memory impairment probability for the user; and providing a clinical decision support recommendation for a provider based on the memory impairment probability.

[0006] In some embodiments, the method further comprises receiving non-audio results of the assessment.

[0007] In some embodiments, the assessment comprises a DCTclock test, a verbal recall test, and a speech metric test.

[0008] In some embodiments, the method further comprises extracting one or more graphomotor feature from a result of the DCTclock test.

[0009] In some embodiments, one or more graphomotor features are selected from Table 4.

[0010] In some embodiments, the acoustic and speech timing features comprise frequency, decibel level, speech duration, reaction time, and / or pause rate.

[0011] In some embodiments, the acoustic features are selected from Table 2 and the speech timing features are selected from Table 3.

[0012] In some embodiments, the machine learning model is a random forest model.

[0013] In some embodiments, the random forest model is a random forest regressor.

[0014] In some embodiments, the result for the one or more multimodal assessments comprises data collected from one or more of a touchscreen, a stylus, a webcam, and / or a microphone.

[0015] In some embodiments, applying the machine learning model comprises applying a series of stacked logistic regressions producing a series of scores associated with features of cognitive function.

[0016] In some embodiments, the series of scores correspond to one or more of drawing efficiency, simple and complex motor function, speed of information processing, and / or visuospatial reasoning.

[0017] In some embodiments, the method further comprises a composite weighted score of the one or more audio or speech timing features.

[0018] In some embodiments, the method further comprises normalizing the composite score by age- or gender-category values.

[0019] In some embodiments, the machine learning model is pretrained with a standard memory assessment to obtain a memory impairment ground truth.

[0020] In some embodiments, the audio feature is automatically time-segmented and transcribed by an automatic speech recognition model.

[0021] In some embodiments, the method further comprises detecting speech patterns based on a fundamental frequency and / or loudness of the audio feature.

[0022] In some embodiments, the one or more assessments include a verbal memory test.

[0023] In some embodiments, the memory impairment probability includes an estimate for one or more subtypes of memory impairment.

[0024] In an alternative embodiment, a system comprises at least one input device; a computing node coupled to the at least one input device and comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a processor of the computing node to perform a method comprising: receiving one or more audio recordings of an interaction by a user with a computing device during an assessment; extracting acoustic and speech timing features from the one or more audio recordings; applying a machine learning model to the acoustic and speech timing features to generate a memory impairment probability for the user; and providing a clinical decision support recommendation for a provider based on the memory impairment probability.

[0025] In an alternative embodiment, a computer program product for determining a cognitive impairment status, the computer program product comprising a computer readable storage medium having program instructions embedded therewith, the program instructions executable by a processor to cause the processor to perform a method comprising: receiving one or more audio recordings of an interaction by a user with a computing device during an assessment; extracting acoustic and speech timing features from the one or more audio recordings; applying a machine learning model to the acoustic and speech timing features to generate a memory impairment probability for the user; and providing a clinical decision support recommendation for a provider based on the memory impairment probability.BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The accompanying drawings, which are incorporated into and constitute a part of this specification, illustrate various exemplary embodiments and together with the description, serve to explain the principles of the disclosed embodiments.

[0027] FIG. 1 is a process diagram of a method for estimating a level of memory impairment in a patient, in accordance with one or more embodiments of this disclosure.

[0028] FIG. 2A is a graph detailing the results of logistic regression models used to predict verbal memory related cognitive impairment, in accordance with one or more embodiments of this disclosure.

[0029] FIG. 2B is a graph detailing the results of logistic regression models used to predict verbal memory related cognitive impairment, in accordance with one or more embodiments of this disclosure.

[0030] FIG. 3A is a graph depicting a waveform over time and estimated speech segments over time, in accordance with one or more embodiments of this disclosure.

[0031] FIG. 3B is an illustration of a DCTclock drawing and the associated scoring graph, in accordance with one or more embodiments of this disclosure.

[0032] FIG. 4 is a flow diagram illustrating an architecture of a voice and speech data processing workflow, in accordance with one or more embodiments of this disclosure.

[0033] FIG. 5 is an example DCRP core cognitive evaluation Web Report, in accordance with one or more embodiments of this disclosure.

[0034] FIG. 6 is an example DCRP Patient Summary Page, in accordance with one or more embodiments of this disclosure.

[0035] FIG. 7 is an example Result as shown on an Admin App, in accordance with one or more embodiments of this disclosure.

[0036] FIG. 8A is an exemplary display of completed patient assessments, in accordance with one or more embodiments of this disclosure.

[0037] FIG. 8B is an exemplary display of a provider report for a patient, in accordance with one or more embodiments of this disclosure.

[0038] FIG. 9A illustrates an example speech assessment, in accordance with one or more embodiments of this disclosure.

[0039] FIG. 9B illustrates an example speech assessment, in accordance with one or more embodiments of this disclosure.

[0040] FIG. 10 is an exemplary computing node.DETAILED DESCRIPTION

[0041] Reference will now be made in detail to the exemplary embodiments of the present disclosure, examples of which are illustrated in the accompanying drawings. Wherever possible, the same reference numbers will be used throughout the drawings to refer to the same or like parts.

[0042] The systems, devices, and methods disclosed herein are described in detail by way of examples and with reference to the figures. The examples discussed herein are examples only and are provided to assist in the explanation of the apparatuses, devices, systems, and methods described herein. None of the features or components shown in the drawings or discussed below should be taken as mandatory for any specific implementation of any of these devices, systems, or methods unless specifically designated as mandatory.

[0043] Also, for any methods described, regardless of whether the method is described in conjunction with a flow diagram, it should be understood that unless otherwise specified or required by context, any explicit or implicit ordering of steps performed in the execution of a method does not imply that those steps must be performed in the order presented but instead may be performed in a different order or in parallel.

[0044] As used herein, the term “exemplary” is used in the sense of “example,” rather than “ideal.” Moreover, the terms “a” and “an” herein do not denote a limitation of quantity, but rather denote the presence of one or more of the referenced items.

[0045] Traditional cognitive screening tools (e.g., Mini Mental State Exam (MMSE), Montreal Cognitive Assessment (MoCA), or MiniCog) lack the sensitivity of detecting cognitive impairment in populations at risk of or at mild stages of cognitive decline. Patients often present with cognitive complaints long after subtle symptoms of cognitive impairment have begun; thus, these traditional tests are often administered too late to catch initial decline, after the onset of overt symptoms, and are often not very sensitive. Additionally, traditional assessments do not easily fit the limited time constraints of primary-care settings, which are frequently at the forefront of cognitive screening, without significant workflow modifications. Cognitive screenings may require administration training to reliably administrate a test and can require expertise to interpret the results and the significance of the findings.

[0046] Embodiments of the present disclosure can capture different domains of cognitive impairment and, particularly, be able to include assessment of the domains primarily affected by the most common neurodegenerative disorders. Some embodiments utilize multimodal digital assessments (voice and speech, written metrics, etc.) to detect memory impairment, a subcomponent of general cognition.

[0047] With the inclusion of a declarative verbal recall task during which speech utterances are digitally recorded and analyzed, speech timing and acoustic voice features may be included in the analysis of cognitive status. Speech and voice feature analysis is feasible,scalable, and cost-effective, especially when compared to comprehensive neuropsychological batteries which can take several hours to complete. Additionally, speech and voice feature analysis does not rely on language contexts (such as semantics, syntax, and pragmatics) which are more difficult to analyze and differ between languages. Through digital sensor capture on a mobile device, speech and voice metrics, and analysis of drawing praxis (including the ability to conceptualize, plan, and organize a task in order to successfully execute a motor skill from beginning to end) can help classifying individuals with cognitive impairments from their cognitively unimpaired (CU) counterparts. Embodiments of the present disclosure include a system and method of linking stylus-based drawing praxis features, voice and speech metrics, and declarative verbal recall accuracy to predict various subtypes of memory performance and potential impairment (working, declarative, procedural, semantic, and prospective memory).

[0048] Embodiments of the present disclosure apply a multimodal approach to classifying cognitive impairment including a protected cognitive screening, graphomotor process, acoustic voice, and speech timing analysis to attempt to classify cognitive impairment. Moreover, some embodiments apply multimodal acoustic voice, speech timing, graphomotor, and declarative verbal recall features and outcomes to predict specific memory impairment subtypes. In contrast, other solutions are likely to rely on only voice or lengthy speech task feature extraction.

[0049] Embodiments of the present disclosure are directed towards a system and method for incorporating graphomotor praxis, speech production, and voice quality metrics captured by digital sensors on a mobile device to predict various broad subcomponents of potential memory impairment. Several digitized assessments examining various subcomponents are incorporated into a customized cognitive assessment and administered via a tablet device, mobile device, or other suitable processor. Verbal responses to memory tasks are recorded and stored in a cloud environment. The recorded audio is time segmented and transcribed using a cloud-based platform to transcribe audible speech. The recall audio is analyzed and acoustic (F0, dB, etc.) and speech timing features are extracted. This analysis allows for the detection of speech via dB and frequency bands where speech is most likely present. The resultant output of the semi-automated analyses may be manually reviewed for accuracy. Speech timing and acoustic voice features obtained from the audio are subsequently included in random forest classifier models to analyze feature contributions to predicting cognitive status and model ability to predict cognitive impairment status. The result is a probabilitymodel and corresponding probability known as the “ Memory Impairment Probability” or MIP, provided as a percentage likelihood score. Using this MIP, predictive values may be extracted for hippocampal dependent memory problems.

[0050] Some embodiments rely on a quick, verbal recall task together with a short cognitive screener to classify cognitive impairment better than voice and speech or cognitive screening alone. Importantly, the algorithm is trained using standardized verbal memory assessment to obtain a memory impairment classification ground truth. Logistic regression classifier models with missing forest imputation are used to incorporate features across multiple modalities and thus, increase cognitive classification accuracy. The model includes age and gender controlled DCTclock graphomotor metrics, DCR verbal recall accuracy, and DCRP acoustic and temporal speech features provided the most accurate means of classifying cognitive impairment on a binary scale (as shown in FIG. 3A-B). The proposed system and method for classification of memory impairment combines these features in a statistical model to provide robust classification accuracy of ADRD memory impairment and is used to provide the MIP.

[0051] In an initial stage of the analysis described above, a patient may be directed to complete a battery of assessments, including a Digital Clock and Recall (DCR) assessment, among others. For example, a DCR assessment may include voice and speech analysis score is added to a scoring algorithm where the speech timing and acoustic voice is added as an addition to a 3 -word recall task. The DCR assessment involves a patient drawing two clocks on a tablet. In particular, in the DCR assessment, a patient is asked to “freeform” draw a clock and then asked to draw a clock with a particular time using the traditional two hands of the clock followed by a verbal recall test. The algorithm tracks the patient’s progress at 120x / second and correlates the tracked data to other existing data related to normal cognitive function and potential cognitive impairment. As another example, a Trails A&B assessment involves asking a patient to connect dots between letters (A-Z) and numbers (1-9 et. al.) that are interspersed and scattered throughout a display screen. The tracked patient responses and lines drawn may be correlated with data related to normal and abnormal cognitive function. As yet another example, a DCTclock analysis involves measuring how a pen or stylus stroke moves as compared to a smooth pen movement during the drawing process. The results of this assessment may be captured and / or included when the pen or stylus is used, regardless of which test / assessment is being completed. The battery of assessments can be conducted on a tablet, smartphone, or any other processing device where the patient is able to respond.Embodiments of the present disclosure use established and proven norms published in literature for iterations of voice and speech testing. It is contemplated that alternative embodiments can incorporate a machine learning process that does not rely on published norms and instead will make use of both (1) standardized threshold-based assessments; and (2) sample-based thresholding for an individual patient determined by the machine learning model.

[0052] For the delayed three-word verbal memory test / assessment, patients are presented with three words to remember. Immediately following encoding of the three-word recall task, patients complete the DCTclock drawing tasks, as described above. Following the completion of the DCTclock drawings (typically requiring 2-3 minutes to complete and thus, acting as the delay for the verbal recall task), patients are asked to recall the 3 words originally encoded in the immediate recall condition.

[0053] Verbal memory has been researched extensively as a means of differentiating and classifying disease populations from healthy individuals. Many of the most common word recall tasks used in clinical practice in the United States are derived from classic neurologic assessments. For example, the Mini-Cog is a brief pencil-and-paper screening tool commonly used to assess cognition in populations with known or suspected cognitive impairment. This tool pairs immediate and delayed word recall with a clock drawing task to assess memory, attention, and executive functioning. Traditionally, the Mini-Cog provides six, 3 -word sets so that when the test is administered at different time points, different word sets can be used to avoid learning and carry-over effects. However, while appropriate for American English speakers, some of the words used in the recall task have less relevance in other languages or are overly complex and are subsequently more difficult to remember. Accordingly, a new, cross-linguistically applicable set of words for use in word recall sets is included in the battery of assessments. In addition to being able to use the same sets of words across languages, the goal was to create a set of words that are commonly occurring (as determined by word frequency) and of similar speech sound (phoneme) complexity. To perform more sophisticated speech and voice production analyses, vowel diversity was a secondary focus of the new word set; recently, researchers have examined specific motor speech features extracted from acoustic signals of vowels in words to differentiate disease populations from healthy individuals. These analyses require a diverse set of vowel productions, including both central (produced with the tongue at rest in the center of the mouth) and non-central (requiring some movement of the tongue to produce). Thus, more canbe inferred about speech motor ability, cognition, and vocal health and quality. Table 5 describes the important features of the original and new word sets in terms of the average normalized frequency of occurrence, phonemic complexity index, and counts of central and non-central vowels.

[0054] The battery of assessments, such as those described above, can include a range of tests designed to characterize the patient’s cognitive ability and physiology. After the patient completes the DCR assessment within the battery, the resulting raw data may be analyzed and objective scoring metrics may be produced to describe various aspects of the feature sets gathered (see Table 5, below) of the patient’s performance during the assessment. Embodiments of the present disclosure can incorporate five metrics / features that correlate with Ap± status: Delayed Recall score, DCTclock Spatial Reasoning composite score for the Copy clock, DCTclock Oscillatory Motion feature score for the Command clock, DCTclock Average Speed feature score for the Copy clock, and DCTclock Maximum Speed feature score for the Copy clock.

[0055] An additional benefit of capturing speech and voice features for inclusion in a multimodal (graphomotor, cognitive, verbal memory, etc.) analysis of brain function, is the unique ability to detect motor impairment specific to corticobulbar (head and neck) function (fine motor control, including basal ganglia and cerebella’s systems, etc.), emotional characteristics (depression, apathy, anxiety, etc.), and anatomical abnormalities associated with structures used to produce speech and voice (muscle tension, polyps, nodules, etc.).

[0056] The multimodal machine learning model, known as the DCRP (DCR plus voice and speech features) for classifying memory impairment using standardized memory impairment assessment provides the MIP. Audio is time segmented and transcribed using an automatic speech recognition (ASR) software platform. The algorithm is applied to word recall audio to extract acoustic features of fundamental frequency (e.g., mean, standard deviation, minimum, maximum, and jitter), loudness (e.g., mean, minimum, maximum, standard deviation, and shimmer). Metrics of speech timing output were calculated according to the automatic speech recognition output.25, 36-38 The algorithm allows for the detection of speech obtained from decibel (dB) and frequency (Hertz; Hz) bands where speech segments are most likely found (30-80 dB and 70-500 Hz threshold bands were used in this analysis). A complete list of the temporal, acoustic and clock drawing features extracted on the platform are listed in Tables 2-4. Speech timing and acoustic voice features obtained from the audio were subsequently included in the DCRP model to classify cognitive impairment status.

[0057] FIG. l is a process flow diagram 100 of an exemplary method for predicting a MIP. For example, the method (e.g., steps 101-104) may be performed by a processor automatically or in response to a request by the user. The method may include one or more of the following steps. In step 101, the method may include receiving one or more audio recordings of an interaction by a user with a computing device during an assessment. In step 102, the method may include extracting acoustic and speech timing features from the one or more audio recordings. In step 103, the method may include applying a machine learning model to the acoustic and speech timing features to generate a memory impairment probability for the user. In step 104, the method may include providing a clinical decision support recommendation for a provider based on the memory impairment probability.

[0058] FIG. 2A details the results of logistic regression models (DCRP, mini mental state exam (MMSE), DCR, DCT, Age + Gender, Gender, and Age) used to predict verbal memory related cognitive impairment (Table 1). FIG. 2B details the results of logistic regression models (DCRP, MMSE, DCR, DCT, Age + Gender, Gender, and Age) used to predict verbal memory related cognitive impairment (Table 1). Area Under the Receiver Operating Characteristic (ROC) Curve (AUC), classification accuracy (Acc), sensitivity (Sens), and specificity (Spec) were used to evaluate the model performance. Sens provides a percentage of impaired individuals correctly identified where Spec represents the percentage of individuals correctly identified as unimpaired. Finally, positive and negative predictive values (PPV & NPV) indicate the proportions of correct positive and negative memory impairment classification (Table 1). Table 1 indicates classifier performance for DCRP, DCR, DCT, and MMSE models in predicting memory impairment determined by age adjusted delayed verbal recall test scores. PPV refers to positive predictive value; NPV refers to negative predictive value; and ACC refers to accuracy.Table 1: Classifier performance

[0059] Results indicate that features included in the DCRP speech and voice model provided the highest classification value (AUC = 0.83; sensitivity = 0.81, specificity = 0.80). This is an improvement over the DCR (AUC = 0.82; sensitivity 0.80, specificity 0.79), and DCTclock of only (AUC = 0.77; sensitivity = 0.75, specificity = 0.70) model feature sets. These results are a further and more balanced improvement in sensitivity and specificity over models using MMSE (AUC = 0.85; sensitivity = 0.80, specificity = 0.72). The DCRP outperforms all single domain and demographics models as predictors of delayed verbal recall-determined memory impairment (Table 1).

[0060] FIG. 3A depicts a graph 300 illustrating the waveform 310 over time and the estimated speech segments 330 based on dB level analysis (y axis) over time (x axis; in seconds). The trace 320 follows fluctuations in dB above the speech threshold.

[0061] FIG. 3B illustrates a sample of DCTclock drawing 340. The drawing is scored in terms of drawing efficiency, simple & complex motor, information processing, and spatial reasoning (as shown in graph 350). This individual was classified as having cognitive impairment.Clock Drawing Analysis

[0062] DCTclock derived features contribute significantly to some DCRP models. The DCTclock component comprises four composite scores (information processing, drawing efficiency, simple and complex motor functions, spatial reasoning), and each of these composite scores are, in turn, made up of a host of individual metrics with standard normative values. Various groupings of subscales can be extracted to interpret the process of conducting a neuropsychological and / or behavioral task. Differences in these composites and subscales can be associated with functional impairment that could be indicative of cognitive impairment, ADRD or another type of brain-related disability. In addition, changes in these composites and subscales could be associated with an individual change in functional status that could be indicative of cognitive decline, ADRD or another type of brain-related disability.Preprocessing and Modeling

[0063] Prior to performing normalization and imputation, an input dataset is split 80-20 (N=754, N=188) into training and test sets, ensuring a balanced composition of impaired and unimpaired participants in the test set. For normalization, the training set is fit andtransformed using a min-max scaler for a range of 0 to 1. The test set is then transformed using the same scaler previously fit to the training set. For imputation, a missForest iterative imputation technique may be used, whereby a random forest regressor may be employed to impute all missing values for each feature. This may be done iteratively to support imputation accuracy. As with normalization, imputation may be fit to the training set, then used to transform both the training and test sets.

[0064] All classification analysis conducted may be performed using logistic regression. DCRP, DCR, and DCT models for both delayed verbal recall and MMSE prediction were subject to hyperparameter tuning, determined using the mean results of a 10-fold cross- validation (10 CV) grid search method on the training set. For smaller feature set models (i.e., Age, Gender, Age and Gender, delayed verbal recall, MMSE), an L2 regularization penalty (Ridge) may be applied, whereas for larger feature set models (i.e., DCRP, DCR, DCT), an elastic net regularization penalty may be used with varying LI ratio values.Statistical Feature Selection

[0065] Features for DCRP (DCR plus speech & voice features), DCR, and DCT models may be pruned first on the basis of relevance and redundancy, then further pruned using a 10- fold cross-validation recursive feature elimination method on the training set. First, features with below 0.15 Pearson correlation with the target may be removed from the feature set. Passing features may then be assessed for any high correlation across all features in the set; any features with correlation above 0.99 with another feature of higher correlation with the target may be subsequently removed as well. The remaining features are then assessed for their rate of missing values; any with a rate above 0.05 may be individually assessed for high correlation with features outside of the set, specifically those with a lower rate of missing values. If other features are discovered to have a correlation above 0.75 with the high missing value feature, said features may then be reincorporated into the feature set. If no other features are found meeting these requirements, and the assessed feature has a rate of missing values in excess of 0.10, then the feature may be removed from the final set.

[0066] FIG. 4 provides a general description and flow of speech input. FIG. 4 is a schematic 400 providing a general description and flow of the processing of speech from intake to processing to output of the MIP. Within the schematic 400, upstream processes 402 and model repository 404 are input to an analytics engine 408. The upstream processes 402 may comprise data received from multimodal assessments, processed for input into theanalytics engine 408 for processing and analysis. The model repository 404 may comprise a number of suitable models trained to analyze the input data. The analytics engine 408 may comprise a segment-specific engine lambda wrapper, which imports between a segmentspecific core engine python module and / or a voice / speech python module. The analytics engine 408 output may be sent to a segment-specific SNS topic and metrics writer Lambda, ultimately terminating as input to other downstream processes 410. The analytics engine 408 may perform analysis on the input data to provide the MIP. Additional metrics for the input multimodal data may be derived from a return metric loop 406. The return metric loop 406 may include sending data from the analytics engine 408 to a preprocessed audio file bucket, along with intermediate files. Data may be read from the bucket and used as input to Praat Engine 412. Alternatively, the return metric loop 406 may be invoked and data may be sent from the analytics engine 408 directly to the Praat Engine 412. The Praat Engine 412 may comprise a parselmouth lambda wrapper and a Praat engine. Data may return from the return metric loop 406 to the analytics engine 408.Memory Impairment Probability in Healthcare Delivery

[0067] The addition of voice and speech metrics to the DCR (DCR plus voice & speech (DCRP)) improves the model’s cognitive-classification capabilities. One of the capabilities enabled by the DCR is the Memory Impairment Probability (MIP). The MIP is a percentage score (0-100%) based on a logistic-regression classifier which incorporates delayed recall performance, speech and voice analytics, and graphomotor features to predict and measure verbal memory and its impairments in clinical and research settings. Predicting a patient’s delayed recall performance (e.g., on delayed verbal recall) is crucial because it reflects a given individual’s verbal episodic memory - the cognitive function most relevant to AD pathophysiology.

[0068] The MIP leverages algorithmically selected voice / speech metrics extracted from the Immediate / Delayed Recall segments and features of the DCTclock to enable the DCR to estimate, with excellent precision (± 3%), the likelihood of episodic verbal memory impairment in a given patient. The following examples illustrate the levels of the likelihood of verbal memory impairment estimated by the MIP:0-50%: Low likelihood of verbal memory impairment51-80%: Moderate likelihood of verbal memory impairment81- -100%: High likelihood of verbal memory impairmentThese probability levels help determine the clinical decision support recommendations for next diagnostic or interventional steps suggested to the PCPs based on the likelihood of the presence of amnestic (likely AD-related) cognitive impairment.

[0069] FIG. 5 is an exemplary DCRP core cognitive evaluation (CCE) web report for a sample patient. A user can view the results within minutes once the assessment has been completed by the patient. The report 500 in FIG. 5 is an example of what it could look like on a Web Portal. The score 501 and performance are displayed with a breakdown of the subscore 502, Clock Drawing 503, and Delayed Recall 504 segments of the DCR, which make up the total score 501. Additional DCR results may be contextually displayed on the right. This is a configurable feature that can be turned on or off for any clinic. The MIP and interpretation may appear to the right of the overall DCR score and performance. The MIP title includes an info icon, and hovering over it reveals a tooltip with additional details about the MIP, including scoring and interpretation cut-offs, brief context on calculation, and references.

[0070] FIG. 6 is an exemplary DCRP patient summary page for the sample patient. The displayed summary 600 includes all of the patient’s most recent assessment scores, including the MIP. A clinician may compare current scores to previous scores. A user may reach this page through the Patient List (by clicking the patients’ icon in the left-side navigation rail), from the web report (by clicking the patient’s name in the breadcrumb at the top of the page), or by clicking on 'View recommended follow-up.' To return to the web report from this screen, the user can click 'View Report' on the right side of the Digital Clock and Recall card.

[0071] FIG. 7 is an exemplary DCRIP Provider Report PDF 700. In the web report 500 (FIG. 5), users can simply click the Provider Report button located in the upper right corner to either download or print the report as shown in FIG. 6. Furthermore, some embodiments of the system include the capability to seamlessly integrate reports for instant export to a clinic's EHR, and it remains conveniently accessible through the Admin App for immediate viewing once the assessment has been completed (FIG. 8A-B). FIG. 8A-B are exemplary views of a patient’s result on an administrative application.

[0072] FIG. 8A illustrates a display in an administrative application of a completed list of patient assessments. Upon selecting “view”, a provider report can be shown for any patient listed. FIG. 8B illustrates an alternative web view display of a provider report PDF for a patient.Memory Impairment in Other Assessments

[0073] Additionally, the MIP score can be derived from all applicable voice and speech assessments. These assessments can be completed across a multitude of device types, ranging from completing in clinic on an iPad or tablet device to completing remotely on a mobile device or computer / laptop. The MIP can be computed and applied to assessments where voice and speech are elicited on the same platform. Similarly, using these assessments helps differentiate between different types of memory. Each of these tasks allows the provider to assess more than episodic declarative and procedural memory components. The MIP lays the groundwork for analyzing Executive Control and Visuconstructional classifications using voice analytics. For example, cross-examining the results of delayed free recall and delayed cued recognition allows us to differentiate types of cognitive impairment.

[0074] FIGs. 9A-B show exemplary speech assessments available on the system platform. Memory Impairment Probability in Pharmaceutical. Research, and Life Sciences

[0075] Within the life sciences industry, there is a unique opportunity to provide value using the MIP score. Screening is a vital component of the research process, and use of the MIP score may allow researchers to quickly identify whether a participant matches the criteria of their study. Similarly, researchers may use this information to track participants overtime to see if any of their participants suffer adverse side effects.

[0076] Table 2 is a list of speech timing features that may be extracted from the recall audio provided by the patient and used as input for the model. Table 3 is a list of acoustic voice features that may be extracted from the recall audio.Table 2: List of speech timing features extracted from recall audio.Table 3: List of acoustic voice features extracted from recall audio.Table 4: List of drawing features obtained from the DCTClock algorithm.Table 5: Exemplary Battery of Assessments

[0077] Referring now to FIG. 10, a schematic of an example of a computing node is shown. Computing node 10 is only one example of a suitable computing node and is not intended to suggest any limitation as to the scope of use or functionality of embodiments described herein. Regardless, computing node 10 is capable of being implemented and / or performing any of the functionality set forth hereinabove.

[0078] In computing node 10 there is a computer system / server 12, which is operational with numerous other general purpose or special purpose computing system environments or configurations. Examples of well-known computing systems, environments, and / or configurations that may be suitable for use with computer system / server 12 include, but are not limited to, personal computer systems, server computer systems, thin clients, thick clients, handheld or laptop devices, multiprocessor systems, microprocessor-based systems, set top boxes, programmable consumer electronics, network PCs, minicomputer systems, mainframe computer systems, and distributed cloud computing environments that include any of the above systems or devices, and the like.

[0079] Computer system / server 12 may be described in the general context of computer system-executable instructions, such as program modules, being executed by a computer system. Generally, program modules may include routines, programs, objects, components, logic, data structures, and so on that perform particular tasks or implement particular abstract data types. Computer system / server 12 may be practiced in distributed cloud computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed cloud computing environment, program modules may be located in both local and remote computer system storage media including memory storage devices.

[0080] As shown in FIG. 10, computer system / server 12 in computing node 10 is shown in the form of a general -purpose computing device. The components of computer system / server 12 may include, but are not limited to, one or more processors or processing units 16, asystem memory 28, and a bus 18 that couples various system components including system memory 28 to processor 16.

[0081] Bus 18 represents one or more of any of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, and a processor or local bus using any of a variety of bus architectures. By way of example, and not limitation, such architectures include Industry Standard Architecture (ISA) bus, Micro Channel Architecture (MCA) bus, Enhanced ISA (EISA) bus, Video Electronics Standards Association (VESA) local bus, Peripheral Component Interconnect (PCI) bus, Peripheral Component Interconnect Express (PCIe), and Advanced Microcontroller Bus Architecture (AMBA).

[0082] Computer system / server 12 typically includes a variety of computer system readable media. Such media may be any available media that is accessible by computer system / server 12, and it includes both volatile and non-volatile media, removable and non-removable media.

[0083] System memory 28 can include computer system readable media in the form of volatile memory, such as random access memory (RAM) 30 and / or cache memory 32. Computer system / server 12 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, storage system 34 can be provided for reading from and writing to a non-removable, non-volatile magnetic media (not shown and typically called a "hard drive"). Although not shown, a magnetic disk drive for reading from and writing to a removable, non-volatile magnetic disk (e.g., a "floppy disk"), and an optical disk drive for reading from or writing to a removable, non-volatile optical disk such as a CD-ROM, DVD-ROM or other optical media can be provided. In such instances, each can be connected to bus 18 by one or more data media interfaces. As will be further depicted and described below, memory 28 may include at least one program product having a set (e.g., at least one) of program modules that are configured to carry out the functions of embodiments of the disclosure.

[0084]

[0007] Program / utility 40, having a set (at least one) of program modules 42, may be stored in memory 28 by way of example, and not limitation, as well as an operating system, one or more application programs, other program modules, and program data. Each of the operating system, one or more application programs, other program modules, and program data or some combination thereof, may include an implementation of a networkingenvironment. Program modules 42 generally carry out the functions and / or methodologies of embodiments as described herein.

[0085] Computer system / server 12 may also communicate with one or more external devices 14 such as a keyboard, a pointing device, a display 24, etc.; one or more devices that enable a user to interact with computer system / server 12; and / or any devices (e.g., network card, modem, etc.) that enable computer system / server 12 to communicate with one or more other computing devices. Such communication can occur via Input / Output (I / O) interfaces 22. Still yet, computer system / server 12 can communicate with one or more networks such as a local area network (LAN), a general wide area network (WAN), and / or a public network (e.g., the Internet) via network adapter 20. As depicted, network adapter 20 communicates with the other components of computer system / server 12 via bus 18. It should be understood that although not shown, other hardware and / or software components could be used in conjunction with computer system / server 12. Examples, include, but are not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data archival storage systems, etc.

[0086] The present disclosure may be embodied as a system, a method, and / or a computer program product. The computer program product may include a computer readable storage medium (or media) having computer readable program instructions thereon for causing a processor to carry out aspects of the present disclosure.

[0087] The computer readable storage medium can be a tangible device that can retain and store instructions for use by an instruction execution device. The computer readable storage medium may be, for example, but is not limited to, an electronic storage device, a magnetic storage device, an optical storage device, an electromagnetic storage device, a semiconductor storage device, or any suitable combination of the foregoing. A non-exhaustive list of more specific examples of the computer readable storage medium includes the following: a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), a static random access memory (SRAM), a portable compact disc read-only memory (CD- ROM), a digital versatile disk (DVD), a memory stick, a floppy disk, a mechanically encoded device such as punch-cards or raised structures in a groove having instructions recorded thereon, and any suitable combination of the foregoing. A computer readable storage medium, as used herein, is not to be construed as being transitory signals per se, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagatingthrough a waveguide or other transmission media (e.g., light pulses passing through a fiberoptic cable), or electrical signals transmitted through a wire.

[0088] Computer readable program instructions described herein can be downloaded to respective computing / processing devices from a computer readable storage medium or to an external computer or external storage device via a network, for example, the Internet, a local area network, a wide area network and / or a wireless network. The network may comprise copper transmission cables, optical transmission fibers, wireless transmission, routers, firewalls, switches, gateway computers and / or edge servers. A network adapter card or network interface in each computing / processing device receives computer readable program instructions from the network and forwards the computer readable program instructions for storage in a computer readable storage medium within the respective computing / processing device.

[0089] Computer readable program instructions for carrying out operations of the present disclosure may be assembler instructions, instruction-set-architecture (ISA) instructions, machine instructions, machine dependent instructions, microcode, firmware instructions, state-setting data, or either source code or object code written in any combination of one or more programming languages, including an object oriented programming language such as Smalltalk, C++ or the like, and conventional procedural programming languages, such as the “C” programming language or similar programming languages. The computer readable program instructions may execute entirely on the user’s computer, partly on the user’s computer, as a stand-alone software package, partly on the user’s computer and partly on a remote computer or entirely on the remote computer or server. In the latter scenario, the remote computer may be connected to the user’s computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or the connection may be made to an external computer (for example, through the Internet using an Internet Service Provider). In some embodiments, electronic circuitry including, for example, programmable logic circuitry, field-programmable gate arrays (FPGA), or programmable logic arrays (PLA) may execute the computer readable program instructions by utilizing state information of the computer readable program instructions to personalize the electronic circuitry, in order to perform aspects of the present disclosure.

[0090] Aspects of the present disclosure are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the disclosure. It will be understood that each block ofthe flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer readable program instructions.

[0091] These computer readable program instructions may be provided to a processor of a general purpose computer, special purpose computer, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions / acts specified in the flowchart and / or block diagram block or blocks. These computer readable program instructions may also be stored in a computer readable storage medium that can direct a computer, a programmable data processing apparatus, and / or other devices to function in a particular manner, such that the computer readable storage medium having instructions stored therein comprises an article of manufacture including instructions which implement aspects of the function / act specified in the flowchart and / or block diagram block or blocks.

[0092] The computer readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other device to produce a computer implemented process, such that the instructions which execute on the computer, other programmable apparatus, or other device implement the functions / acts specified in the flowchart and / or block diagram block or blocks.

[0093] The flowchart and block diagrams in the Figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of instructions, which comprises one or more executable instructions for implementing the specified logical function(s). In some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams and / or flowchart illustration, and combinations of blocks in the block diagrams and / or flowchart illustration, can be implemented by special purpose hardware-based systems that perform the specified functions or acts or carry out combinations of special purpose hardware and computer instructions.

[0094] The descriptions of the various embodiments of the present disclosure have been presented for purposes of illustration, but are not intended to be exhaustive or limited to the embodiments disclosed. Many modifications and variations will be apparent to those of ordinary skill in the art without departing from the scope and spirit of the described embodiments. The terminology used herein was chosen to best explain the principles of the embodiments, the practical application or technical improvement over technologies found in the marketplace, or to enable others of ordinary skill in the art to understand the embodiments disclosed herein.

Claims

What is claimed:

1. A method, comprising: receiving one or more audio recordings of an interaction by a user with a computing device during an assessment; extracting acoustic and speech timing features from the one or more audio recordings; applying a machine learning model to the acoustic and speech timing features to generate a memory impairment probability for the user; and providing a clinical decision support recommendation for a provider based on the memory impairment probability.

2. The method of claim 1, further comprising receiving non-audio results of the assessment.

3. The method of claim 1, wherein the assessment comprises a DCTclock test, a verbal recall test, and a speech metric test.

4. The method of claim 3, further comprising extracting one or more graphomotor feature from a result of the DCTclock test.

5. The method of claim 4, wherein the one or more graphomotor feature is selected from Table 4.

6. The method of claim 1, wherein the acoustic and speech timing features comprise frequency, decibel level, speech duration, reaction time, and / or pause rate.

7. The method of claim 1, wherein the acoustic features are selected from Table 2 and the speech timing features are selected from Table 3.

8. The method of claim 1, wherein the machine learning model is a random forest model.

9. The method of claim 8, wherein the random forest model is a random forest regressor.

10. The method of claim 1, wherein the result for the one or more multimodal assessments comprises data collected from one or more of a touchscreen, a stylus, a webcam, and / or a microphone.

11. The method of claim 1, wherein applying the machine learning model comprises applying a series of stacked logistic regressions producing a series of scores associated with features of cognitive function.

12. The method of claim 11, wherein the series of scores correspond to one or more of drawing efficiency, simple and complex motor function, speed of information processing, and / or visuospatial reasoning.

13. The method of claim 1, further comprising a composite score weighted combination of the one or more audio or speech timing features.

14. The method of claim 1, further comprising normalizing the composite score by age- or gender-category values.

15. The method of claim 1, wherein the machine learning model is pretrained with a standard memory assessment to obtain a memory impairment ground truth.

16. The method of claim 1, wherein the audio feature is automatically time-segmented and transcribed by an automatic speech recognition model.

17. The method of claim 16, further comprising detecting speech patterns based on a fundamental frequency and / or loudness of the audio feature.

18. The method of claim 1, wherein the one or more assessments include a verbal memory test.

19. The method of claim 1, wherein the estimate of memory impairment includes an estimate for one or more subtypes of memory impairment.

20. A system comprising: at least one input device; a computing node coupled to the at least one input device and comprising a computer readable storage medium having program instructions embodied therewith, the program instructions executable by a processor of the computing node to perform a method comprising: receiving one or more audio recordings of an interaction by a user with a computing device during an assessment;extracting acoustic and speech timing features from the one or more audio recordings; applying a machine learning model to the acoustic and speech timing features to generate a memory impairment probability for the user; and providing a clinical decision support recommendation for a provider based on the memory impairment probability.

21. A computer program product for determining a cognitive impairment status, the computer program product comprising a computer readable storage medium having program instructions embedded therewith, the program instructions executable by a processor to cause the processor to perform a method comprising: receiving one or more audio recordings of an interaction by a user with a computing device during an assessment; extracting acoustic and speech timing features from the one or more audio recordings; applying a machine learning model to the acoustic and speech timing features to generate a memory impairment probability for the user; and providing a clinical decision support recommendation for a provider based on the memory impairment probability.

Images