RESUME SEMINAR ICoLiST

RESUME SEMINAR ICoLiST
Anissafa Atul Khusna- Offering B

Presenter : Mario Rosario Guarracino (University of Cassino)
The Theory : Machine Learning and Artificial Intelligence in bioimaging Application
Overview
 Artificial intelligence and machine learning methdos (Basic concepts)
 Basic problems in computer visions (Image restoration, object, detection, image
classification, image segmentation)
 Biomaging applications and available software
 Issues, open problems and conclusions
Introduction
o Human intelligence has the ability to adapt previous knowledge to new situations and to
recognize meaning in patterns
o Artificial intelligence (Al) aims at replicating these abilities in non-human agents (ABII)
or in creating new ones (AGII)
o Machine learning extracts useful features from large sets of data to make predictions or
decisions on unseen data
o We don’t write anymore software to solve problems, rafther we write programs to train
computers to selve problems
The rise of Al
 Although Neural Netwroks (NNs) were envisioned since the 1950s significant
performances in pattern recognition tasks were only achieved in the late 1980s
 While inference with trained netwroks is generally fast, the training process can take
hours to days
 The success of NNs in image recognition is related to the increase in the power of
Graphcila Processing Units (GPUs)
Deep learning
 In early 2010s. Deep Learning (DL) became increasingly prominent as a tool for image
classifications
 In 2012 the first GPU-enabled NN called AlexNet vastly outperformed the competition at
the ImageNet image classifications challenge a seminal breakthrough for the Al filed
Basic NN architectures
Layers are made up off NODES, which take one of mc weighted input connections and produece
an output connection. They are organized into layers to comprise a network. To progressively
extract higher-level features from data, machine learning models use multi-layer structure called
neural network, consisting of successive finction compositions.
Basic DNN aechitectures
CNN is mostly used for image classifications, image recognition problems. In its methodology
the whole image is scanned with filters. In the literature 1x1, 3x3, and 5x5 filter sizes are mostly
used in most of the CNN architectures, there are different types of layers convolutional pooling
(average or maximum), fully connected layers
Biomaging
Bimaging refers to methods that non-invasively visualize biological processes
 Structural informations (2D or 3D) of tissues, cells and molecules
 Temporal information, e.g cell movement, changes in cell shape
 Single cell
Biomaging has the potential to expand our knowledge of biological mechanism under diseases
and accelerate the process of drug discovery
Biomaging and Al
DNNs have expanded in many disciplines, and have significantly improved biomedical image
anlaysis. Their applications include automates, accurate classification and segmentation of cell
images, extraction of structures from label-free microscopy imaging (artificial labeliling) , and
image restoration (e.g denoising and resolution enhancement). Although DL in microscopy is yet
tp become widely available, the current growth in research efforts will soon fundamentally
change how imaging data is analysed and how microscopy is carried out.
Basic problems in Computer Vision
 Classification (classify the main object category within an image)
 Object detection (identify the object category and locate its position using bounding box)
 Semantic segmentation (separate an image into regions (segments) according to object
category)
 Instance segmentation (separate ab image into regions (segments) according to object
instance)
Object classification and detection
 Goal (recognize and assign identities to relevant objects on a microscopy images)
 Issues (touching cells, inhomogeneous background noise, large variations in cell sizes
and shapes)
 Applications (cell nuclei detection and indentifications ; spotting intranuclear particles;
quantitative analysis)
Objectivity, reliability, and validity
 Manual annotation of features in microscopic image with a low signal-to-noise ratio is
subjective and prone to errors
 Training DL models on subjective annotations and unrepresentative samples may be
instable or yiels biased models
 In turn, these models may unreliably detect biological effects
 A pipeline integrating data annotation, ground truth estimation, and model training which
produces unbiased and reproducible results is still to come
Open problems
 Biomaging arguably faces the biggest challenges in automating visual image
interpretation tasks, due to :
i. The lack of standard imaging protocols
ii. The high variability of experimental conditions
iii. The hug volume of the data produced (on the order of terabytes of image data in a
single experiment)
 Biomaging requires expensive haedware and expensive software
 The integration of multiple microscopy techniques on the seme sa,ple has to be more
investigated to overcome the proper limits of each technique lack of training data
 The success of omics imaging in the analysis of biomedical images might provide more
insight in available data
Conclusions
 Machine Learning and Al are driving forces behind bioimaging processing and analysis
 A wide range of ready-to -use software tools are available to perform many different
tasks
 Among the many limiting factors, the lack of sufficient computational power can prevent
a complete insight in our experiments
 A tight collaboration between BIOimaging and bioMAGING communities can provide
opportunities from both
Pengalaman pertama dalam mendengar dan menerima materi dalam Bahasa Inggris Full
Sejujurnya ini adalah pengalaman pertama saya dalam mendapatkan materi yang
penyampaiannya penuh menggunakan Bahasa Inggiris. Jujur saja saya merasa kesusahan dalam
memahai materinya karena kemampuan listening saya beserta kemampuan Bahasa Inggris saya
kurang bagus, tapi saya merasa senang mendapatkan materi ini karena menambahkan
pengalaman baru bagi saya dan memotivasi saya untuk lebih giat lagi dalam belajar Bahasa
Inggris.
(To be honest, this is my first experience in getting theory that is fully delivered in English.
Honestly, I find it difficult to understand the theory because my listening skills and English skills
are not good, but I feel happy to get this theory because it adds new experience for me and
motivates me to study English more actively)

RESUME SEMINAR ICoLiST

Uploaded by

Copyright:

Available Formats

RESUME SEMINAR ICoLiST

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

RESUME SEMINAR ICoLiST

Uploaded by

Copyright:

Available Formats

RESUME SEMINAR ICoLiST

Anissafa Atul Khusna- Offering B

You might also like