website/Q at main · jwuillou8/website · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
---
layout:      project
title:       Neurotrophic Labs
date:        1 Mar 2021
image:
  path:       /assets/img/past_projects/jwimage-light@0,25x.jpg
  srcset:
    1920w:   /assets/img/past_projects/jwimage-light.jpeg
    960w:    /assets/img/past_projects/jwimage-light@0,5x.jpg
    480w:    /assets/img/past_projects/jwimage-light@0,25x.jpg
caption:     NL is a medical technology allowing MD's to design patient and disease profiles and get correlations based on real datasets in  real time.
description: >
  NL is a technology allowing MD's to design scenarii/hypotheses and get data based evaluations in real time.
links:
  - title:   Prototype
    url:     http://neurotrophic-labs-live-system.herokuapp.com
  #- title:   Slides #Community Report
  #  url:     https://drive.google.com/file/d/1iba4ccoW8jVP8kRXH5AvfI575r9ui6Vr/view?usp=sharing # https://faxi.shinyapps.io/NEXT/
  #- title:   Video
  #  url:     https://drive.google.com/file/d/0BwkQZAFeYw2ZUWVfejczQjkzTUE/view?usp=sharing
featured:    false
---

# Neurotrophic Labs

With Jeremy Martinez, a californian MD,
we have called neurotrophic Labs a collaboration aiming at
developing innovative solutions for MD's.

NL is a flexible technological solution helping MD’s to take data based decisions in real time.

With this work, we are trying to support Doctors decision making, which should impact positively on individuals health and finance.

## App:
We have built a prototype processing 5 millions patient histories from the medicare research dataset.
This architecture demonstrates that MDs can model simple scenarios and get basic correlations and insights from real data,  very quickly.

We list below the main aspects of the approach:
####  Easy design of scenarios & hypotheses
A combination of symbolic algebra and neurocomputing techniques allows for a seamless modelling of patients.
The app allows with auto completion to enter several profile and diseases combinations. The database can be queried against and returns similar profiles. As an example  ($\{\}$ indicates lists):

##### $$ patient\, \,  profile \times medical \, \, history \times \{ diseases \} \longrightarrow  \{ correlations \}$$
Our UX also allows to model scenarios, alternatives and evaluate them quickly and in parallel.
##### $$ \{patient\, \,  profiles \times medical \, \, history\} \times \{ diseases \} \longrightarrow  \{ correlations \}$$


#### Live Data based evaluation
We could perform the operations described above, on the whole data set described above within ~1 second on 8 CPU cores.

#### Online Learning
The user can train the app, editing and creating patient profiles to teach the app via examples ($\{\}$ indicates lists):
##### $$ \{ patient\, \,  profiles\, examples\, 1\, +\, medical \, \, histories\, 1 \} \longrightarrow  \{diseases\, 1\}$$
##### $$ \{ patient\, \,  profiles\, examples\, 2\, +\, medical \, \, histories\, 2\} \longrightarrow  \{ diseases \, 2\}$$
##### $$\dots $$
This mechanism allows for MDs to train the app, allowing to start with small datasets and to gradually learn.

#### Neuromorphic ”Engine”
Patients histories are modeled as binary fingerprints, allowing for
fuzzy matchings either by profile or by disease.

Aside of the speed and flexibility, the other strength of the approach is the complete robust against false negatives
(the system will have perfect matches against critical diseases).

## Team:

We are experienced consultants covering both medical and AI/technical spaces.

Links to Jeremy website and projects: