10.05.2021
Volker Bäcker, JeanBernard Fiche,
Cedric Hassen Khodja, Francesco Pedaci
What is bioimage analysis?
How is it done without machine
learning?
What is machine learning?
How is bioimage analysis done
with machine learning?
“The extraction of information from digital images in the context of biological research”
feature  σ=3.5  σ=7.0  σ=10. 

variance  
sobel 
Machine learning algorithms build a mathematical model of sample data, known as ”training data”, in order to make predictions or decisions without being explicitly programmed to perform the task.





ML algorithm implements a
mathematical model with a number
of model parameters
given the training data,
find parameter values that minimize
the prediction error
Training Data:
Femur length (cm)  Height (cm) 

45  153 
44  168 
44  177 
47  180 
44  171 
50  168 
estimate body height f(x) given the femur length x.
model: f(x) = ω_{1} + ω_{2} × x
parameter of the model:
ω_{1} and ω_{2}
find parameters
ω1, ω2
so that error
between
training data
and model
is minimal
find the minimum of
the loss function
by using gradient
descent
f(x) = ω_{1} + ω_{2} × x
ω_{1} = 131.13cm
ω_{2} = 0.87
f(55cm) = 131.13cm + 0.87 × 55cm
f(55cm) = 179.42cm
Supervised or Unsupervised?
Classification or Regression?
A machine learning method
Unsupervised
Classification
Clustering
Group objects in a way that
objects in the same cluster are
more similar to each other
than to objects in other clusters
Partition the feature
space into kclusters
Each featurevector
belongs to the cluster
with nearest mean
Classification of unknown data:
calculate the feature vector
assign it to the cluster
with the nearest mean
Training phase:
randomly select a number of
feature vectors
run the kmeans clustering on
the selected feature vectors
the resulting means are the
classifier