Analysis and Classification of Biomedical Images From Low Level processing to Decision Support

1
Analysis and Classification of Biomedical
ImagesFrom Low Level processing to Decision
Support

• Sara Colantonio, Davide Moroni, Gabriele Pieri,
  Ovidio Salvetti
• Signals and Images Lab
• Institute of Information Sciences and
  Technologies (ISTI)
• Italian National Research Council (CNR)
• Pisa, Italy

2
Outline

• Image Restoration and Enhancement
• Image Segmentation
• Shape Modeling
• Possible topics of interests for ECSON
• Shape Description and Retrieval
• Knowledge Discovery
• Decision Support

3
Enhancement Restoration

• Aim
• To improve and repair the visual content and
  appearance of biomedical images (low level
  processing)
• To prepare medical images for further processing
  steps (e.g. segmentation, feature extraction,
  classification)
• Methods Convolution Filters, Edge-preserving
  smoothing, PDE-based filters
• Applications
• General pre-conditioning of biomedical images
  (Echocardiography, Magnetic Resonance Imaging,
  Microscopy,)
• Mesh Smoothing for extraction of noisy features

4
Enhancement Restoration Anisotropic Diffusion
Filter
Original
Enhanced
5
Image Segmentation

• Aim to divide images in meaningful sub-regions
  for the analysis of relevant substructures
• Methods Watershed, Active Contours, Level Sets,
  Fuzzy-Neural Methods
• Applications
• Object recognition and classification, Motion
  detection, 3D image analysis, Deformable
  structures modeling
• Scene Understanding, Content-based Image
  Retrieval, MM Annotation
• Diagnostic imaging
• Industrial quality control

6
Image Segmentation Sample Applications

• Medical Imaging

7
Shape Modeling

• Aim to provide methods for representing
  faithfully but compactly objects shape
• Methods Differential geometric properties of
  manifolds Topological properties Level Set
  Methods Registration theory
• Applications
• Biomedical Diagnosis
• Left Ventricle (LV) modeling for the analysis of
  the heart dynamics
• 3D neurofunctional region mapping and brain
  deformation
• Surveillance
• Active Tracking
• Industrial Inspections
• Combustion Instability Characterization, Ceramic
  Tiles Quality Control

8
Shape Modeling a sample application
Deformation Pattern wall thickness
Original Data
Model normalization for modal analysis
Reconstruction and Registration
9
Outline

• Image Restoration and Enhancement
• Image Segmentation
• Shape Modeling
• Possible topics of interest for ECSON
• Shape description and retrieval
• Knowledge Discovery
• Decision Support

10
An idea for a structured contribution
From Shapes to Therapy Planning
11
An idea for a structured contribution
From Shapes to Therapy Planning
12
Basic Shape Retrieval Goal

• Find 3D models with similar shape

3D Query
Best Match(es)
3D Database
13
Extended Shape Retrieval Goal

• Find 3D or 3D1 models with similar shape in a
  database endowed with semantic annotations

Car
3D or 3D1 Query
Best Match(es)
Chair
3D Database Semantics
14
Shape Retrieval in Medical Imaging

• How to build such DB?

Shape Storing
Shape Integration
Shape Annotation
Shape Description
15
Shape Description

• Shape Descriptor
• Structured abstraction of a 3D model
• Capturing salient shape information

3D or 3D1 Query
ShapeDescriptor
Best Match(es)
16
Methods for Shape Description

• Geometrical features
• Many methods
• Volumetric Representations, Shape Distributions,
  Template Matching Parameters, Light fields,..
• High descriptive power

Angular Bins
Model
EGI

• Topological features
• High discriminative power
• Weak robustness w.r.t. small perturbation
• Difficult to distinguish truly features from noise

• How to bring together such powers?

17
Shape Annotation

• Add semantics to the low level description
• E.g. the model refers to a heart left ventricle
  with some ischemic impairments in the apical
  regions
• In a semi-automatic/manual way
• Structure the semantic concepts in a suitable,
  machine-readable way
• Encode relations among the various involved
  concepts
• E.g. the Heart Left Ventricle is a part of the
  Heart
• Ontologies are currently employed to solve these
  tasks

18
Shape Storing

• Define a database structure containing both
• Data
• Metadata (descriptors semantic annotations)
• The structure should allow for
• Similarity searches
• KDD procedures
• Non-trivial problems should be addressed
• Large data amount
• Heterogeneous data
• Efficiency

19
Shape Integration

• Integrate the realm of shapes in the problem
  context

• Gain support for
• More Powerful KDD
• Case-based reasoning
• Effective Decision Support

3D Model
Functional Imaging
Demographics / History
Planning
LAB
20
From Shapes to Therapy Planning
21
Knowledge Discovery in Databases

• AIM of KDD the non trivial extraction of
  implicit, novel and potentially useful
  information from large amounts of data

The KDD Process
22
KDD Mission
Implicit relations among data to be transformed
in valid knowledge
Discovery
ANNOTATED SHAPE DBs
Adaptive Learning methods for clustering,
classification regression
23
From Shapes to Therapy Planning
24
Decision Support Service

• AIM of a DSS to assist decision making processes
  by combining multimedia information,
  sophisticated analytical models and tools and
  user-friendly interface
• Methods encoding of high-level, specialized
  medical knowledge elicited from clinical experts
  and extracted from clinical guidelines into a
  Knowledge Base which is used by an Inferential
  Reasoning process able to supply relevant
  suggestions when needed

25
DSS Development

• Useful also for annotation purposes

Elicitation of relations among concepts
Common Terminology
Elicitation of concepts
Routine procedures description
Encoded by using ontology concepts
Elicitation of procedures
Encoding in suitable formalism
KDD
26
Our Experience

• Development of a Knowledge-based DSS for
  supporting the early diagnosis and
  medical-clinical management of heart failure
  within elderly population the HEARTFAID Project

27
The Ontologies
28
The Base of Rules
Example of natural language elicitation If a
patient has Left Ventricle Ejection Fraction lt
40 and he is asymptomatic and is assuming ACE
Inhibitors and he had a myocardial infarction
then a suggestion for the doctor is to give the
patient Betablockers
29
DSS Suggestions
Care Personalization
Therapy Planning
DSS Suggestion?
DSS Suggestion
30
Bibliographical References (1/2)

• Chiarugi F., Colantonio S., Emmanouilidou D.,
  Moroni D., Salvetti O . - Decision Support in
  Heart Failure through ECG and Echocardiography
  Processing, submitted to Artficial Intelligence
  in Medicine, 2008.
• Barcaro U., Moroni D., Salvetti O. - Automatic
  computation of left ventricle ejection fraction
  from dynamic ultrasound images. In Pattern
  Recognition and Image Analysis, vol. 18 (2) pp.
  351 358., 2008.
• Colantonio S., Martinelli M., Salvetti O.,
  Gurevich I. B., Trusova Y. - Cell image analysis
  ontology. In Pattern Recognition and Image
  Analysis, vol. 18 (2) pp. 332 341, 2008.
• Moroni D., Perner P., Salvetti O. - A general
  approach to shape characterization for biomedical
  problems. In Advances in Mass Data Analysis of
  Signals and Images, pp. 136 145, (Lecture Notes
  in Artificial Intelligence, vol. 4826). Berlin
  Heidelberg Springer, 2007. 
• Chiarugi F., Colantonio S., Emmanouilidou D.,
  Moroni D., Salvetti O. - Biomedical signal and
  image processing for decision support in heart
  failure. In MDA 2008 - Advances in Mass Data
  Analysis of Images and Signals (Leipzig,
  Germany, 14 July 2008). Proceedings, pp. 38 - 51.
  (Lecture Notes in Computer Science, vol. 5108).
  Springer, 2008.
• Moroni D., Salvetti M., Salvetti O. - Multi-scale
  representation and persistency for shape
  description. In MDA 2008 - Advances in Mass Data
  Analysis of Images and Signals (Leipzig, Germany,
  14 July 2008). Proceedings, pp. 123 - 138.
  (Lecture Notes in Computer Science, vol. 5108).
  Springer, 2008.   
• Colantonio S., Conforti D., Martinelli M., Moroni
  D., Perticone F., Salvetti O., Sciacqua A. - An
  intelligent and integrated platform for
  supporting the management of chronic heart
  failure patients. In CinC 2008 - Computers in
  Cardiology 2008 (Bologna, 14-17 Settembre 2008).
  Proceedings, pp. 897 - 900. (Computers in
  Cardiology, vol. 35). Computers in Cardiology,
  2008.

31
Bibliographical References (2/2)

• Chiarugi F., Colantonio S., Emmanouilidou D.,
  Moroni D., Perticone F., Sciacqua A., Salvetti O.
  - ECG and echocardiography processing for
  decision support in heart failure. In CinC 2008
  - Computers in Cardiology 2008 (Bologna, 14-17
  Settembre 2008). Proceedings, pp. 649 - 652.
  (Computers in Cardiology, vol. 35). Computers in
  Cardiology, 2008.
• Colantonio S., Salvetti O., Tampucci M. - An
  infrastructure for mining medical multimedia
  data. In ICDM 2008 - Advances in Data Mining.
  Medical Applications, E-Commerce, Marketing, and
  Theoretical Aspects. (Leipzig, Germany, 16-18
  July 2008). Proceedings, pp. 102 - 113. Petra
  Perner (ed.). (Lecture Notes in Computer Science,
  vol. 5077). Springer, 2008.
• Colantonio S., Gurevich I. B., Salvetti O. -
  Automatic fuzzy-neural based segmentation of
  microscopic cell images. In Advances in Mass
  Data Analysis of Signals and Images , pp. 115 -
  127. Perner Petra, Salvetti Ovidio (eds.).
  (Lecture Notes in Artificial Intelligence, vol.
  4826). Berlin / Heidelberg Springer, 2007. 
• Colantonio S., Salvetti O., Sartucci F. -
  Automatic recognition and classification of
  cerebral microemboli in ultrasound images. In
  Pattern Recognition and Image Analysis, vol. 15
  (2) pp. 532-535. Mank-Hayka/Interperiodica, 2005.
• Di Bona S., Lutzemberger L., Salvetti O. - A
  simulation model for analyzing brain structures
  deformations. In Physics in Medicine and
  Biology, Vol. 48 n. 24 (2003), p. 4001-4022. 
   ISTI-2003-A0-10 - Di Bona S., Niemann H., Pieri
  G., Salvetti O. - Brain volumes characterization
  using hierarchical neural networks. In
  Artificial Intelligence in Medicine 28, n. 3
  (2003), p. 307-322. Elsevier, 2003.