Machine differentiation of abnormalities in bioelectromagnetic fields
Abnormalities in electromagnetic fields in the heart, brain, and stomach, among other organs and tissues of the human body, can be indicative of serious health conditions. Described herein are methods, software, systems and devices for detecting the presence of an abnormality in an organ or tissue of a subject by analysis of the electromagnetic fields generated by the organ or tissue.
1. A diagnostic device configured to detect a cardiac abnormality of a heart of an individual, comprising:
(i) an electromagnetic field sensor configured to sense an electromagnetic field measurement associated with the heart of the individual, wherein the electromagnetic field sensor comprises an optically pumped magnetometer (OPM), a fluxgate, or a superconducting quantum interference device (SQUID);
(ii) a processor operably coupled to the electromagnetic field sensor; and
(iii) a non-transitory computer-readable storage medium encoded with software comprising a trained machine learning software module, wherein the trained machine learning software module comprises a deep neural network (DNN) configured to process an electromagnetic field measurement of the individual to reconstruct a reference spatiotemporal representation for the individual comprising a spatiotemporal representation of heart-related electromagnetic field (EMF) signal segments that is representative of an absence of the cardiac abnormality, wherein the software is executable by the processor and causes the processor to:
(a) receive the electromagnetic field measurement from the electromagnetic field sensor;
(b) generate a test spatiotemporal representation for the individual using the electromagnetic field measurement, wherein the test spatiotemporal representation comprises at least two members selected from the group consisting of a butterfly plot, a spatiotemporal representation of magnetic activation of the heart from base to apex, and a spatiotemporal representation of magnetic activation of the heart from left to right;
(c) process the electromagnetic field measurement using the trained machine learning software module to reconstruct the reference spatiotemporal representation for the individual comprising the spatiotemporal representation of heart-related EMF signal segments that is representative of the absence of the cardiac abnormality;
(d) use a hypothesis function to determine a difference between the test spatiotemporal representation for the individual and the reference spatiotemporal representation for the individual; and
(e) detect the cardiac abnormality of the heart of the individual based on the difference determined in (d).
2. The device of claim 1 , further comprising a plurality of electromagnetic field sensors configured to be positioned at a plurality of different positions relative to a body of the individual to sense a plurality of electromagnetic field measurements associated with the heart of the individual.
3. The device of claim 1 , further comprising a housing containing the processor, wherein the electromagnetic field sensor is hard-connected to the housing.
4. The device of claim 1 , wherein the trained machine learning software module is trained using a plurality of electromagnetic field values sensed from a plurality of individuals.
5. The device of claim 4 , wherein the trained machine learning software module is trained using a plurality of health data values associated with the plurality of individuals.
6. The device of claim 4 , wherein the plurality of electromagnetic field values sensed from the plurality of individuals comprises heart-related data.
7. The device of claim 1 , wherein the trained machine learning software module is trained using one or more of demographic data, medical image data, and clinical data associated with one or more individuals.
8. The device of claim 1 , wherein the software is executable by the processor and causes the processor to further determine a therapy for treating the individual based on the cardiac abnormality.
9. The device of claim 1 , wherein the DNN comprises a deep convolutional neural network (CNN), a deep dilated CNN, a deep fully-connected neural network, or a deep recurrent neural network (RNN).
10. The device of claim 1 , wherein the test spatiotemporal representation generated in (b) comprises the butterfly plot, the spatiotemporal representation of magnetic activation of the heart from base to apex, and the spatiotemporal representation of magnetic activation of the heart from left to right.
11. The device of claim 1 , wherein an abnormal medical diagnosis for the individual is determined when the difference determined in (d) is greater than or equal to a threshold value, and wherein a normal medical diagnosis for the individual is determined when the difference determined in (d) is less than the threshold value.
12. The device of claim 1 , wherein the electromagnetic field sensor comprises the OPM.
13. The device of claim 1 , wherein the software is executable by the processor and causes the processor to further localize an anatomical region of the heart of the individual associated with the cardiac abnormality.
14. The device of claim 1 , wherein the cardiac abnormality of the heart of the individual comprises cardiac ischemia.
15. The device of claim 1 , wherein the cardiac abnormality of the heart of the individual comprises congestive heart failure.