LOADING ...

Your browser does not support WebGL.
You can download one of these popular web browsers.

Device Development:
1) Get access to the device data
2) Understand how to analyze the data (eg. Fast-Fourier-Transform, time/frequency/energy)
3) Create method to display data which indicates relative ranges of these changes
4) Perform rudimentary pattern recognition to identify periodicity of signal features
5) Pilot or limited study: collect data on healthy normal vs. patients with known pathology
6) Determine features representative of variations of signal patterns & assign variables
7) Analyze data to determine features unique to pathology and ranges of variations
8) Create Algorithm based upon variables and their ranges discovered above
9) Assign variables as input nodes to Neural Network, output is score: 0=Normal, 1=Abnormal
10) Perform Machine Learning/Deep Learning/AI to determine Sensitivity & Specificity
11) Depending on level of accuracy, perform assessment of Level-Of-Importance of variables
12) Expand topology of Neural Network as necessary, rerun Machine Learning, score results
13) Expand clinical trial, perhaps separating groups of subjects into age or other categories
14) Score each group and perform analysis, create data charts indicating accuracy of each group
15) Possibly include all clinical data and rerun Machine Learning to determine enhanced score
16) Publish results of entire study

* Imaging: VIS-NIR-IR-UV, stereo/3D, Global-shutter, QR target tracking
- Volumetric rendering: DICOM Voxal, 3d mesh Tomography
* Sensors: spectral, acoustic, pressure, chemical (including polymer), radiological
* COTS chip-level and FPGA/CPU programmable
* Micro-endoscopes: illumination controller with real-time compensation
* Modulation controllers

* Programming of the device FPGA/CPU board
- Arduino, custom boards: Python, C++/C#/Java/JavaScript
* Methods to get data from device
- Direct wired: USB/COM port/Audio-data
- Wireless: Bluetooth (Virtual COM port)/WiFi/Cell-data

* Mobile Apps (iPhone, iPad, Android)
- Multi-platform analysis-&-results capable App (Windows, Mac, WebBrowsers)
* Cloud Development
- Programming of server-side analysis, database updates
- Big data mining, data pattern discovery analytics, distributed computing
- Notifications
-- Mobile SMS
-- Web-portal/WebGL/Web Apps, website updates, permissions based
--- Login to view patient results

Algorithm Development:
1) Get access to the device data
2) Understand how to analyze the data (eg. Fast-Fourier-Transform, time/frequency/energy)
3) Create method to display data which indicates relative ranges of these changes
4) Perform rudimentary pattern recognition to identify periodicity of signal features
5) Pilot or limited study: collect data on healthy normal vs. patients with known pathology
6) Determine features representative of variations of signal patterns & assign variables
7) Analyze data to determine features unique to pathology and ranges of variations
8) Create Algorithm based upon variables and their ranges discovered above
9) Assign variables as input nodes to Neural Network, output is score: 0=Normal, 1=Abnormal
10) Perform Machine Learning/Deep Learning/AI to determine Sensitivity & Specificity
11) Depending on level of accuracy, perform assessment of Level-Of-Importance of variables
12) Expand topology of Neural Network as necessary, rerun Machine Learning, score results
13) Expand clinical trial, perhaps separating groups of subjects into age or other categories
14) Score each group and perform analysis, create data charts indicating accuracy of each group
15) Possibly include all clinical data and rerun Machine Learning to determine enhanced score
16) Publish results of entire study

* Create, research or obtain a thorough description of the problem
* Analyze the problem, develop full understanding from knowledge base
* Develop hypothesis supported by published findings
* Create a set of Rules based-on knowledge base
* Define a set of variables representing known data features
* Develop the high-level "Expert System" version of the Algorithm, generally a math formula
* Test and refine algorithm with data (if available)
* Review algorithm for portions that benefit from optimization from Machine Learning

* Create Spectral Analysis, 2D spectrographs & 3D spectrograms
* Determine S/N ratio, noise floor, ceiling, scaling ranges
* Define energies at specific peak times, frequency ranges & temporal features
* Pattern Analysis:
- Perform rudimentary pattern recognition, pattern matching
- Discover key features that differentiate normal vs. abnormal
- Analyze frequency width, frequency center point of peak energy, curve shapes (slopes), temporal shift, frequency shift, rate shift, rhythm shift

* Create variables for each identified key feature
- Create math function:
-- Feature variables (v[n]) and weights (w[n])
-- Evaluate Level Of Importance (LOI) for each pair
-- Example: (v1 * w1) + (v2 * v2) + … (v[n] * v[n]) = 0..1
-- Resulting score: 0 = normal, 1 = abnormal

Machine Learning:
1) Get access to the device data
2) Understand how to analyze the data (eg. Fast-Fourier-Transform, time/frequency/energy)
3) Create method to display data which indicates relative ranges of these changes
4) Perform rudimentary pattern recognition to identify periodicity of signal features
5) Pilot or limited study: collect data on healthy normal vs. patients with known pathology
6) Determine features representative of variations of signal patterns & assign variables
7) Analyze data to determine features unique to pathology and ranges of variations
8) Create Algorithm based upon variables and their ranges discovered above
9) Assign variables as input nodes to Neural Network, output is score: 0=Normal, 1=Abnormal
10) Perform Machine Learning/Deep Learning/AI to determine Sensitivity & Specificity
11) Depending on level of accuracy, perform assessment of Level-Of-Importance of variables
12) Expand topology of Neural Network as necessary, rerun Machine Learning, score results
13) Expand clinical trial, perhaps separating groups of subjects into age or other categories
14) Score each group and perform analysis, create data charts indicating accuracy of each group
15) Possibly include all clinical data and rerun Machine Learning to determine enhanced score
16) Publish results of entire study

* Neural Networks
- Determine number of input nodes and hidden layers
-- Input nodes may be sensor inputs, or variable values
- Determine Level Of Importance (LOI) of inputs
- Determine if dynamic topology or static number of layers is sufficient

* Adaptive Models
- Evolutionary Computation, Genetic Algorithms
- Continual Self Adaptation (Dynamic model)
-- Fast results: scores available during ongoing optimization
- Deep Learning on Big Data, continual
- Distributed Computing using multiple PCs for faster deep learning

* Dynamic Algorithm
- Continually learns & adapts to new data, further optimizing accuracy
- Allows for expansion of more finely granulated categories of data
-- Patient age groups can be further narrowed/refined
-- Patients with other conditions can be considered in separate group(s)
- Allows more accurate versions of algorithm to be certified and introduced to replace previous
* Locked Algorithm
- Specificity and Sensitivity locked for certifications
- Generally used to run on tablets/mobiles as version with specific accuracy

About:
ABSTRACT: Timothy James Hays
Currently: Chief Technical Officer at Cerenetex, Inc.
Cognitive Scientist, Medical Device Devloper
Machine Learning (AI), Neural Networks, Algorithms
40+ Years Software Development, 15+ Years Medical Algorithms

Background:
Software development with primary focus: AI Algorithms

Methods:
Evolutionary Computation, Genetic Algorithms, Artificial Object Neural Networks, Distributed Deep Machine Learning, Signal Analysis for spectrums: Acoustic, UV-VIS-NIR-IR, vapor chemical mass-spectroscopy

Results:
Over 100 past projects for Science, Medical, Education & Entertainment

Conclusions:
Perfect for Sensor Data Analysis and advanced Signal Processing with 3D Spectrograms

* Clinical trials (HIPAA compliant)
- Pilot study (Phase I)
-- Acquire data from known pathology
-- Compare against healthy volunteers
-- Discover features in signal data toward developing algorithm
-- Prototype tests to determine changes necessary during Phase II
- Limited Clinical Trial (Phase II)
-- Algorithm developed and machine learning applied on final data
-- Expected to prove viability of Phase II Minimum Viable Product
-- Assess final commercial product features
Extended Clinical Trial (Phase III)
-- Verification & Validation of data against locked AI Algorithm
-- Results typically establish Sensitivity & Specificity for commercialized device

* Patents

- Medical
NON - INVASIVE SYSTEMS AND METHODS FOR THE IMPROVED EVALUATION OF PATIENTS SUFFERING FROM UNDIAGNOSED HEADACHES

- AI, Neural Networks, Genetic Algorithms
Video Game Characters have Evolved Traits

- Games
Method for presenting a virtual reality environment for an interaction