Documenti di Didattica
Documenti di Professioni
Documenti di Cultura
line 1: 4th Given Name Surname line 1: 5th Given Name Surname
line 2: dept. name of organization line 2: dept. name of organization
(of Affiliation) (of Affiliation)
line 3: name of organization line 3: name of organization
(of Affiliation) (of Affiliation)
line 4: City, Country line 4: City, Country
line 5: email address or ORCID line 5: email address or ORCID
2
There are a lot of terms being used so let’s visualize them one
by one.
FIG 2.2: Input is converted to feature map.
CNN is also computationally efficient. It uses special We continue like this and aggregate the convolution results in
convolution and pooling operations and performs parameter the feature map (Fig 2.4).
sharing. This enables CNN models to run on any device,
making them universally attractive.
B. ARCHITECTURE OF CNN
All CNN models follow a similar architecture (Fig 2.1). There
is an input image that we’re working with. We perform a
series
FIG 2.1: CNN architecture
C. CONVOLUTION
The main building block of CNN is the convolutional layer.
FIG 2.4: Input to feature map using filter
Convolution is a mathematical operation to merge two sets of
Information. In our case the convolution is applied on the input
data using convolution filter to produce a feature map (fig 2.2).
4
D. POOLING
After a convolution operation we usually perform pooling to
reduce the dimensionality. This enables us to reduce the
number of parameters, which both shortens the training time
and combats overfitting. Pooling layers downsample each
feature map independently, reducing the height and width,
keeping the depth intact.
E. LATEST PROGRESS
Machines have been built up so as to directly capture the cell
image from the blood samples.
F. KEY HIGHLIGHTS
Disease prediction in probability.
Dataset is verified as it was released by the US
FIG 2.5: Pooling window government medical department.
in the window. Similar to a convolution, we specify the This algorithm was been used previously for leaf
window size and stride. disease prediction, we have converted it for malaria
detection.
Here is the result of max pooling using a 2x2 window and The time taken by this algorithm to run on current
stride 2. Each color denotes a different window. Since both the dataset has been reduced from 24 to 20 hours.
window size and stride are 2, the windows are not overlapping. Processing is done without using GPU.
Note that this window and stride configuration halves the size
of the feature map. This is the main use case of pooling, III. PROBLEM STATEMENT
downsampling the feature map while keeping the important
information There are a handful of methods to test for malaria:
Blood smears
Now let’s work out the feature map dimensions before and Antigen testing
after pooling. If the input to the pooling layer has the
dimensionality 32x32x10, using the same pooling parameters The blood smear process can be visualized by:
described above, the result will be a 16x16x10 feature map. First, a blood sample is taken from a patient and then
Both the height and width of the feature map are halved, but placed on a slide.
the depth doesn’t change because pooling works independently The sample is stained with a contrasting agent to help
on each depth slice the input. highlight malaria parasites in red blood cells.
A clinician then examines the slide under a
By halving the height and the width, we reduced the number of microscope and manually counts the number of red
FIG 2.6: Reduce number of weights blood cells that are infected.
weights to 1/4 of the input (Fig 2.6). Considering that we In order to help make malaria testing a faster process in the
typically deal with millions of weights in CNN architectures, field, scientists and researchers have developed antigen tests
this reduction is a pretty big deal. for Rapid Diagnosis Testing (RDT).
An ideal solution would, therefore, need to combine
DISADVANTAGE the speed of RDTs with the accuracy of microscopy.
5
that the proposed method can gain 98%, 93.3%, 98.65% and
90.33% of sensitivity, specificity, prediction value positive and
prediction value negative, respectively.
IX. CONCLUSION
true positives
Recall=
true positives+ false negatives
…(i)
RECALL = 4796/4796+160
=0.9677
true positives
Precision=
true positives+ false positives
…(ii) FIG 8.1: Epoch vs. Loss/Accuracy
PRECISION = 4796/4796+161
=0.9675
2∗precision∗recall
F 1= …
precision+recall
(iii)
F1 = 2*(0.9675*0.9677)/(0.9675+0.9677)
= 0.9675
tp+ tn
Accuracy = …
tp+tn+fp +fn
(iv)
ACCURACY = 0.97
8