ACTIVITY CLASSIFICATION AND COMMUNICATION SYSTEM FOR WEARABLE MEDICAL DEVICE

Abstract

A method for transmitting activity information from a wearable medical device (101) to a patient monitoring system (105), wherein the method comprises generating an activity data packet, wherein the activity data packet comprises at least a first activity field indicative of a recent activity and a second activity field indicative of a past activity, and transmitting the activity data packet from the wearable medical device (101) to the patient monitoring system (105).

Claims

1. A method for transmitting activity information from a wearable medical device to a patient monitoring system, wherein the method comprises: generating an activity data packet, wherein the activity data packet comprises a least a first activity field indicative of a recent activity and a second activity field indicative of a past activity; and transmitting the activity data packet from the wearable medical device to the patient monitoring system; and wherein the first activity field and the second activity field each comprise: a first activity subfield indicative of an activity type and a second activity subfield indicative of a certainty of the activity type.

2. The method according to claim 1, wherein the activity data packet comprises a header field, wherein the header field comprises: a first header subfield indicative of a time range represented by an activity field; and/or a second header subfield indicative of the number of activity fields comprised within the activity data packet.

3. The method according to claim 1, wherein transmitting the activity data packet from the wearable medical device to the patient monitoring system comprises acknowledge-free transmitting of the activity data package.

4. The method according to claim 1, wherein generating the data packet comprises interleaving at least one interleaved activity field based on reasoning and wherein the interleaved activity field comprises a first activity subfield indicative of an activity type and a second activity subfield indicative of a certainty of the activity type.

5. The method according to claim 1, wherein generating the activity data packet comprises interleaving at least one interleaved activity field based on reasoning and wherein the interleaved activity field comprises a first activity subfield indicative of an event and a second activity subfield indicative of a certainty of the event.

6. The method according to claim 1, wherein generating the activity data packet comprises interleaving at least one interleaved activity field based on reasoning and wherein the interleaved activity field comprises a first activity subfield indicative of an attribute of an activity type indicated in the first activity field and a second activity subfield indicative of a value of the attribute.

7. The method according to claim 1, wherein the wearable medical device comprises an accelerometer; and wherein deriving the value of the first activity subfield comprises detecting the orientation of the wearable medical device based on raw accelerometer sensor data.

8. The method according to claim 1, wherein the wearable medical device comprises an accelerometer and wherein deriving the value of the first activity subfield comprises detecting the acceleration magnitude of the wearable medical device within a short time frame based on the raw sensor data.

9. The method according to claim 1, wherein the wearable medical device comprises a sensor system and wherein deriving the value of the first activity subfield comprises detecting a periodicity and/or a cadence in the raw sensor data.

10. The method according to claim 1, wherein deriving an activity type and/or reasoning comprises classifying using a naive Bayes model.

11. The method according to claim 1, wherein deriving an activity type and/or reasoning comprises classifying based on a machine learning algorithm performing a quadratic discriminant analysis or a linear discriminant analysis.

12. The method according to claim 1, wherein deriving an activity type and/or reasoning comprises classifying with a machine learning algorithm using a neural network.

13. A wearable medical device comprising: a sensor system; a classifier for generating an activity data packet, wherein the activity data packet comprises at least: a first activity field indicative of a recent activity and a second activity field indicative of a past activity; and; a device communication unit for transmitting the activity data packet to a patient monitoring system; and wherein the first activity field and the second activity field each comprise: a first activity subfield indicative of an activity type and a second activity subfield indicative of a certainty of the activity type.

14. A patient monitoring system comprising: a system communication unit for receiving and processing of an activity data packet comprising at least: a first activity field indicative of a recent activity and a second activity field indicative of a past activity; and wherein the first activity field and the second activity field each comprise: a first activity subfield indicative of an activity type and a second activity subfield indicative of a certainty of the activity type.

Description

BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In the drawing

[0049] FIG.1 shows a wearable medical device and a patient monitoring system.

DETAILED DESCRIPTION OF EMBODIMENTS

[0050] FIG. 1 shows a wearable medical device 101 comprising a sensor system 102, more specifically a 3-axis accelerometer, a classifier 103 and a device communication unit 104. Activity data packets may be sent from the wearable medical device 101 to a patient monitoring system 105 over a wireless connection 106.

[0051] The wireless connection 106 may be based on packet-switched near-field radio. A typical data packet for a packet-switched near-field radio may have the structure according to table 1. Therein, PHY relates to the physical layer, MAC to media access control, i.e. the data link layer, NET to the network layer and MIC to a message integrity code according to the Open Systems Interconnection model (OSI).

TABLE-US-00001 TABLE 1 PHY MAC NET Application Payload MIC max 90 bytes max 132 bytes

[0052] According to the exemplary structure pursuant to table 1, the application payload, i.e. the effective payload, is at maximum 90 bytes. Pursuant to an embodiment of the method, a format of the application payload, i.e. an activity data packet, may have the format as shown in table 2.

TABLE-US-00002 TABLE 2 T N AF(1) . . . AF(n − 1) AF(n) AF(n + 1) . . . AF(N) max 86 bytes

[0053] As shown, the activity data packet may comprise header information T, N and activity fields AF. The field T in the header refers to a time range the activity information corresponds to (e.g., 1 second, 1 minute, etc.). The field N may indicate the number of activity fields comprised within the activity data packet. The first activity field AF(1) may correspond to the most recent time period. The activity data packet may comprise additional activity fields AF(n) relating to past time periods.

[0054] Table 3 shows an exemplary activity data packet in bitstream syntax, wherein ActivityClassifierPacket relates to an activity data packet, acFieldType to the field T, acNumberofFields to the field N and acActivityType as well as acActivityStrength to an activity field as described hereinbefore.

TABLE-US-00003 TABLE 3 Syntax No. of bytes Mnemonic ActivityClassifierPacket( ) { acFieldType; 1 u_8 acNumberOfFields; 1 u_8 for (n = 0; n < acNumberOfFields; n++) { acActivityType[n]; 1 u_8 acActivityStrength[n]; 1 u_8 } }

[0055] The first field of the activity data packet, acFieldType, is an indicator, which describes the time range each activity field (acActivityType, acActivityStrength) represents. The time ranges may be coded as proposed in table 4 (the suffix h here and in the following indicates hexadecimal notation).

TABLE-US-00004 TABLE 4 Hex 00h 01h 02h 03h 04h 05h 06h 07h 08h 09h 0Ah 0Bh 0Ch 0Dh 0Eh 0Fh t [s] 0 0.5 1 2 5 8 10 30 60 90 120 300 600 900 1800 3600

[0056] As shown, the first 16 possible values represent activity field durations in seconds. The selection 00h may represent a special case, in which the activity fields all correspond to simultaneous current activities. The further 240 possible values are reserved for later use and allow for an adaption of the activity data packet for future developments. Accordingly, the respective remaining four bits are masked.

[0057] The number N of activity fields may be limited to 42. The activity fields may, as shown in table 3, be split into a first subfield acActivityType and a second subfield acActivityStrength. Each subfield may be a one byte subfield. It is possible to represent 256 activity types by one byte, which may be split into groups including activities, events and attributes, for example according to table 5.

TABLE-US-00005 TABLE 5 Code Activity/Event/Attribute 00h Device on table 01h Device loose 02h Unknown/Uncertain 11h Lying 12h Rolling 13h Restless lying 20h Seizure/jerks 21h Shivering 22h Sleeping 30h Sit 31h Eat 32h Drink 33h Communicating/Talking 34h Reading/Using a tablet computer 35h Physical exercise 50h Standing 51h Walking 52h Walking with crutches 53h Walking with assistance 54h Walking with support (walker) 55h Walking using rail 56h Walking in treadmill 60h Wheelchair (self-propelled) 61h Wheelchair (pushed by someone) 90h Walking stairs down 91h Walking stairs up 92h Elevator A0h Toileting A1h Showering A2h Gymnastics/Rehabilitation B0h Fall from walking B1h Fall from bed C0h Bed exit C1h Early bed exit C2h Bed entry CAh Collision D0h Step rate D1h Step regularity D2h Stability/balance D3h Symmetry in ambulating D4h Impact in steps E0h Awakeness FFh Place

[0058] In table 5, activity types coded from 00h to AFh may represent activities, those from B0h to CFh events and those from D0h to FFh attributes. The second subfield of an activity field relating to an activity may indicate the certainty that the activity has been correctly identified. Attributes may refer to the preceding activity. For example, if an activity field indicates the activity type “walking”, the following activity field may provide attribute for this “walking” activity such as the step rate. Accordingly, the activity field may indicate in the first subfield the type of the attribute, e.g. the step rate, and in the second subfield a value for said attribute, e.g. the step rate in steps per minute. A special attribute coded FFh may indicate the place or location where the preceding activity has taken place. In this case, the second subfield may indicate the place according to the location codes shown in table 6.

TABLE-US-00006 TABLE 6 Hex 00h 01h 02h 03h 04h 05h 06h 07h Place Bed Patient Hall- Waiting Treat- Cafe Toilet Shower room way room ment room

[0059] In an example according to table 7, an activity data packet may comprise six activity fields (N=06h) each representing a time period of 10 seconds (T=06h), of which only the first three activity fields are shown.

TABLE-US-00007 TABLE 7 T N AF(1, T) AF(1, S) AF(2, T) AF(2, S) AF(3, T) AF(3, S) . . . 06h 06h 50h 52h 30h D4h 11h f0h . . .

[0060] The first activity field AF(1) indicates that the most recent activity has been standing (AF(1,T)=50h) with a certainty of 52h, which has been preceded by sitting (AF(2,T)=30h), which has been preceded by lying (AF(3,T)=11h).

[0061] Based on reasoning, it may be determined that in going from lying to walking the patient must have performed a bed exit and that the sitting took place in bed because it occurred right before a bed exit.

[0062] Hence, as shown in table 8, the activity data packet for the next time period may be augmented with this additional information.

TABLE-US-00008 TABLE 8 T N AF(1, T) AF(1, S) AF(2, T) AF(2, S) AF(3, T) AF(3, S) AF(4, T) AF(4, S) AF(5, T) AF(5, S) . . . 06h 06h 51h 80h 50h 52h C0h DDh 30h d4h FFh 00h . . .

[0063] As shown, the most recent activity changed from standing (AF(2,T)=50h) to walking (AF(1,T)=51h). Moreover, interleaved activity fields AF(3) and AF(5) have been interleaved indicating the bed exit event (AF(3,T)=C0h) and that sitting took place in bed (AF(5,T)=FFh, AF(F,S)=00h).

[0064] Said additional information may be valuable for a better treatment of the patient.

[0065] In an embodiment, a number of numerical features may be computed from the raw accelerometer data provided by the sensor system 102. These features may, for example, relate to the orientation of the wearable medical device, an acceleration magnitude within a short time frame, a periodicity or cadence in the raw data or other physical time-series. A classifier based on a machine learning algorithm may be used to derive an activity type from the numerical features. More particularly, the classifier may be based on a naive Bayes classifier principle and be trained using a collection of manually annotated activity data from different activity types from several patients (e.g., lying in bed probably asleep, moving actively in bed, sitting, standing, walking, ambulating using a wheelchair).

[0066] A classifier using the naive Bayes classifier principle may provide a likelihood score for each of the activity types. The winning activity type may be selected as activity type corresponding to the current time frame. Typically, the duration of a time-frame of an activity type analysis may be one second.