Automated reliable
2D/3D Computer Vision for Tumour Detection
from CT/MRI scans

Summary from LinkedIn

I am very excited to announce my new position as CEO and Principal Engineer in my own new company.
As such, my implemented automated tumour detection (for livers) has achieved an accuracy of 99.7% (ROC AUC).

Such an achievement does not come easy.
(*) One could say that this is because of special hardware. Well, not at all, an "ordinary" GPU with 48 GB is fine and will do the trick.
(Still, results are generated nearly in real time within less than a minute max including saving, report, statistics etc.)

(*) One could say that this is because of the used network and approach. Well, yes - an already award winning approach (not me) has been considerably refined.

(*) One could say that this is because of data and the number of samples - bingo, this is it.
I have invented and successfully applied a method of how to multiply given (annotated) samples (CT or MRI) by a factor of appr. 40 to 60.
Mathematically sound and safe.
For example, your AI-package will not only train on 100 samples but then on 4000 to 6000 samples. This makes a huge difference and significantly improves accuracy.

Other innovative methods are applied within the process of the data-generator in order to fight/prevent overfitting.

Tumour-detection will always be based on highly unbalanced datasets.
There is something - also mathematically - better than a commonly used Dice-coefficient.

The current tumour-detection for livers will be extended to multi-class detection with benign, malign - and yes - also metastatic anomalies.
Also detecting metastases is a big step towards an accepted clinical application. {Regards livers there is no "defined" classification system such as Gleason-grades for prostate}.

Apart from highest expertise in 2D/3D-Computer Vision for biomedical - or industrial - applications I would be happy to contribute to your success with components mentioned above - or software/algorithms/methods in general for Computer Vision, Artificial Intelligence or Optimization for Industrial Applications.

Looking forward to hearing from you.

PS: Interestingly enough 2D/3D Computer Vision for Autonomous Driving helped a lot for innovative solutions for 2D/3D tumour-detection...