ShapE, Motion and Body composition to Anatomy – SEMBA

SEMBA leverages multi-modal data to learn the correlations between the internal and external observations of the human body.
Registration techniques are used to bring into correspondence the multiple modalities and machine learning techniques are used to learn their correlations and predictive models.

The first result is a statistical anatomical model of the human, accounting for the adipose, muscle and bone tissue.
The model will be made openly available for research purposes.

The second result is a collection of methods allowing to infer the quantities and distribution of adipose, muscle and bone tissue from dynamic external measurements, as well as their evaluation in terms of accuracy, complexity and ease of use.

SEMBA’s original goals do not include the development of a final product, and rather seek the comprehension and modelling of the relations between the internal and external measurements.

Scientific production includes publications at conferences and journals as well as open source software.

Knowing the distribution of adipose, muscle, and bone tissue in the human body anatomy is crucial in the diagnosis of diseases such as type II diabetes, the planning of therapies, the guidance of the therapeutic gestures and the assessment of a therapy’s outcome. Our current knowledge of the adipose, muscle, and bone tissues is based on the internal imaging of in-vivo patients. Examples of these imaging techniques are Computed Tomography (CT), Dual Energy X-Ray Absorption (DXA) and Magnetic Resonance Imaging (MRI). While these modalities allow accurate measurements of the inside of the body, they involve heavy and expensive equipment as well as time consuming procedures.

External dynamic measurements can be acquired with optical scanning equipment, e.g. cameras or depth sensors. These are becoming cheaper and of higher spatial and temporal resolution, allowing accurate scanning of living, moving bodies. While optically based imaging techniques provide accurate, highly dynamic (60 frames per second) reconstructions of the surface shape, they only capture the shape and appearance of the human surface.

The SEMBA project will research the relations between the observations obtained with these two modalities (internal and external scanning), and provide innovative methods in order to infer the internal measurements from the dynamic external ones. Precisely, SEMBA has two main objectives. The first objective is to create a statistical anatomic model of the human body, accounting for the distribution of adipose, muscle and bone tissue. The second objective of SEMBA is to develop methods to obtain a subject-specific instance of the anatomic model from external dynamic measurements of the human body.