Advanced Quantum ML Algorithms For Prospective Digital Health Developers

Advanced Quantum ML Algorithms For Prospective Digital Health Developers August 10th, 2023 #95
Presenter: ChemicalQDevice CEO Kevin Kawchak

Part I) Cirq and TensorFlow Quantum TFQ Libraries
a) Pulse programming
i) Trapezoidal pulse to the coupler between two adjacent qubits
ii) Args: hold time, coupling, rise time, padding time
iii) Detuning: qubit zero detune, qubit one detune

b) Algorithm transpilation
i) Cirq.transformers: transforms input circuit to output circuit
ii) Cirq.AbstractCircuit: base class for Circuit-like objects
iii) Cirq.TwoQubitCompilationTargetGateset: Abstract base class

c) Implementation of quantum ML into other models
i) Moment by moment simulation: simulate_moment_steps
ii) Symbol values, non-fixed values: resolved at runtime
iii) Mixed state simulations: cirq.DensityMatrixSimulator

Part II) Existing Machine Learning Models
a) Apple Core ML Models
i) Images: FCRN-DepthPrediction, Resnet50 Classification, DeeplabV3 Segmentation
ii) Images: YOLOv3 Object Detection, PoseNet Pose Estimation, more
iii) Text : BERT-SQuAD Stanford “Question Answering” Dataset

Part III) Healthcare/Fitness Developer Platforms
a) Google Fit, "Activity goals to improve health", based on Java and others
b) Samsung Tizen "Open source operating system based on Linux", C/C++
c) Apple HealthKit "Provides a central repository for health and fitness data", Swift

Part IV) Developer Health SDK Resources
a) Android Fit Samples
b) Samsung Tizen Examples
c) Compare Fitbit to Healthkit (Need Token)
d) Apple WWDC 2021 Watch Workout App
e) Unscary Apple HealthKit
f) Fetch User's Step Data with HealthKit

Note: The Following Section Is For Educational Purposes Only

Part V) Introduction Machine FDA Learning
a) Good Machine Learning Practice (GMLP)
i) Principles “to help promote safe, effective, and high-quality medical devices” that use AI/ML

b) SaMD Pre-Specifications (SPS)
i) Software as a Medical Device manufacturer’s anticipated modifications to “performance” or
“inputs,” or changes related to the “intended use”

c) Algorithm Change Protocol (ACP)
i) Methods a manufacturer follows “achieve and appropriately control the risks” delineating from the SPS
Part VI) Machine Learning and FDA

a) Probable Benefit Greater Than Probable Risk
i) Demonstrated by establishing “the absence of unreasonable risk of illness or injury associated with the
use of the device for its intended uses and conditions of use”

b) Predicate Device
i) 510(k) approval demonstrates that device is “safe and effective by proving substantial equivalence to a
legally marketed device (predicate device)

c) FDA and Neurology AI/MI
i) See Also ChemicalQDevice FDA PRISMA Studies: Neurology, Segmentation, and Reconstruction

Part VII) FDA AI/ML Example Apple Inc.
a) 510(K) K213971 06/03/2022 Atrial Fibrillation History Feature Cardiovascular “QDB”
i) “Atrial Fibrillation (AFib) History Feature” OTC software-only intended for users ≥ 22 years old w/o AFib
ii) “Opportunistically analyzes pulse rate data to identify episodes of irregular heart rhythms suggestive of
AFib”. Not intended to provide irregular rhythm notifications or replace traditional AFib methods
iii) Intended for use with the Apple Watch and the Health app on iPhone

b) 510(K) K212516 10/26/2021 Photoplethysmograph “QDB” analysis software for OTC use
i) “IRNF 2.0 person-level sensitivity (88.6%) and specificity (99.3%)” both demonstrated to be non-inferior
to those of the predicate device. Predicate Manufacturer Apple Inc. IRNF App De Novo (DEN180042)
ii) Over 2500 subjects, 3 million pulse rate recordings, Machine Learning = Convolutional Neural Network
iii) IRNF 2.0 is substantially equivalent to IRNF, “no differences in technological or performance
characteristics that raise new questions of safety and effectiveness”

c) De Novo DEN180042 8/8/2018 Photoplethysmograph “QDB” analysis software for OTC use
i) The Irregular Rhythm Notification Feature cannot detect heart attacks. Does not constantly look for AFib
and “should not be relied on as a continuous monitor”
ii) User motion, temperature, blood flow can affect device, don’t introduce to strong electromagnetic fields
iii) In 2018, measured a one-minute beat-to-beat tachogram every 4 hours, depending on user activity
See also: 8/6/2018 NIH review on wearable Photoplethysmography (PPG) Technology/FDA Code QDB

Тэги:

#fda #apple #google #samsung #tizen #healthkit #googlefit #medical #device #medicaldevice #cirq #tensorflowquantum #quantum #quantumcomputing #quantummachinelearning #quantumml

Скачать видео