DeepLesion (base of KT-pictures)

2018: The world's largest database of KT-pictures with the mentioned pathologies is available to everyone

In July, 2018 the database of KT-pictures DeepLesion became publicly available. The command which developed it from clinical center of National Institute of health care of the USA notes that DeepLesion is the largest database of KT-images available for all comers. It included more than 32,000 annotated pictures from more than 10,000 clinical cases. Such extensive annotated collections of radiological images are absolutely necessary for deep learning of algorithms on the basis of artificial intelligence which should learn to distinguish different defeats.

One of the images which are stored in DeepLesion base

Ke Yan, the leading researcher of a command, and the radiologist of clinical center Ronald Summers noted that the new database has the huge potential for use in the field of the automated detection (CADe) and diagnostics (CADx) of defeats. DeepLesion differs from the majority of other databases of medical images in the fact that they are directed to identification only of one type of defeats whereas DeepLesion is rather various and extensive to be used for diagnostics of multiple changes.

When studying KT-images radiologists of clinical center measure and note clinically significant finds using "electronic tabs" which can be rather difficult and include shooters, guide lines and the text. All these notes are necessary precisely to describe location and the amount of defeat and to give the chance to other specialists to reveal growth of a tumor. Now these notes became a basis of the DeepLesion database which integrated the annotated KT-pictures of different pathology of any localization there are nodal new growths in lungs, liver tumors, the increased lymph nodes, etc.

The lack of such database served as the main obstacle for development of more universal programs, however the DeepLesion command already created the universal detector of pathological changes and considers that in the future it can be used for screening.^[1]

Notes