Image analysis

Original and result image, Pneumonia positive 83%

NOTE! Image from Stanford - just an example….

What is tested

All 14 classifications of NIHCC Dataset are tested with following accuracies:

…note this is bogus data - accuracies are somewhere in the ballpark…

Classification	Accuracy
Atelectasis	0.80
Cardiomegaly	0.88
Effusion	0.86
Infiltration	0.73
Mass	0.85
Nodule	0.76
Pneumonia	0.75
Pneumothorax	0.88
Consolidation	0.79
Edema	0.80
Emphysema	0.87
Fibrosis	0.76
Pleural Thickening	0.79
Hernia	0.85

Training data

Models are tested and rained against NIHCC dataset, which consists of over 100000 classified chest X-Ray images. Before you use this API for professional use, careful tests with images from your X-Ray devices should be done.

The NIHCC Dataset is available here.

If you can provide us classified and anonymized X-Ray image datasets, we are more than happy to test our model against them - and if allowed, use that dataset also for improving our models.

Pricing

Price is 0.07€ / request.

Technical stuff

For image based analysis we use best of breed components: Tensorflow, Torch, Amazon AI. We are not focused to certain AI technology - instead we follow the research and technological development very closely and try to pick the best algorithms and tools for the problems we are trying to solve. When new and better solutions come along, we will change our underlying stack - but try to keep our API the same.

Our current Chest X-Ray analysis is trained with NIH dataset…. and for