Medical Data Analysis - Lecture 1

• The lecture is titled "Health Information Management and Medical Data Analysis".
• L01 focuses on the Analysis of Medical Data.
• The presenter is Prof. Dr. Mohammed Elmogy from Mansoura University, Faculty of Computers & Information, Dept. of Information Technology.
• The lecture was presented on 09. October 2022.

Lecture Outline

• The Analysis of Medical Data
• Data Analysis in the Clinical Workflow
• Data Analysis Tools
• Case Studies of CAD systems

Introduction - Medical Image Analysis

• It's a rewarding field involving investigating, developing, and applying methods of image processing, computer vision, and pattern recognition.
• Applying image analysis in medicine can save time or increase the reliability of interpretation tasks for medical procedures.
• Adequate selection, correct application, and sufficient validation of the method are important.

Difference Between Medical Images and Others

• Analysis questions can be based on detection or tracking task in a photograph.
• Medical images appearance of the depicted object is from absorption of x-rays, not light reflection.
• Medical images show distributions of physical features from the human body that would otherwise be inaccessible.
• The analysis of medical images has specific expectations which influences the analysis type and algorithm requirements.

Analysis of Medical Images

• Object shape and appearance variance can be characteristic for some evaluation leading to a need to be captured and studied.
• It is difficult to compare information with reality because there are few noninvasive methods to verify information gained from the pictures.
• Delineation, restoration, enhancement, and registration are important for fusing images from different sources.
• Classification, reconstruction of 3D information, and tracking are important in medical image analysis.

Different Imaging Techniques

• Different imaging techniques produce mappings of several physical attributes.
• The goal is to subject several physical attributes in various ways to inspection.

Computer-assisted Analysis of Medical Images

• Computer-assisted analysis supports the expert in decision tasks.
• Delineation of an object requires solving a segmentation task.
• Detection of an object requires solving a classification task.
• Comparison of object appearance from pictures at different times/modalities requires registration task.

Tasks in Medical Image Analysis

• Different methodology and validation techniques are required based on the task.
• Examples tasks are detection, delineation, and registration.
• Detection focuses on finding aneurysms.
• Delineation focuses on finding heart boundaries.
• Registration focuses on combining 3D DSA with projection DSA.

General Analysis Questions

• How the analysis fits into the clinical routine is an important question to consider.
• Strategies to develop an efficient and effective analysis tool should be addressed.
• Identify necessary information to solve the analysis task and how to acquire it.
• Determine how the quality of the analysis method can be tested
• Determine the tools can be used to minimize effort for the analysis task.

Some Analysis Problems

• Each analysis task differs due to organ/pathology, imaging system, parameters, and equipment.
• Finding a quick solution is often prioritized over fitting the solution into the workflow.
• Method development can be separated from regular practice.
• There is a bias towards experimenting with new methods.

Data Analysis in the Clinical Workflow

• Cohort Study in Epidemiology:
  • It's a new field using images in large cohort studies.
  • Its purpose is to study public health on a grand scale.
  • Analysis methods should be largely automatic and performed on large datasets.
  • Radiomics, a new strategy for large image repositories, adopts analysis techniques from genomics for radiological images.
• Clinical Study:
  • Images are analyzed outside a clinic routine to understand or confirm findings.
  • Images are often copied to a workstation.
  • Analysis methods are implemented on the workstation, and results are kept there
  • Data transfer to the workstation must be organized and checked.
• Diagnosis Support (CA detection, CA diagnosis):
  • Single cases may be several studies analyzed on a medical workstation.
  • Transfer of images is often done by other software, it has to be part of the analysis method for access to the images.
  • Diagnosis support systems often involve user interaction to accept or reject analysis results.
• Analysis in Treatment Planning Precedes Treatment:
  • Methods have to account for acceptable task completion time.
  • Treatment planning results are the input for the treatment by a medical expert.
  • Generated results should be well documented.
  • Input into a treatment module may happen automatically.
  • The transfer interface has to be specified.
• Image Analysis for Computer-Assisted Surgery Is Time-Critical:
  • Fast transfer to the system on which the analysis method is implemented has to be organized.
  • Time constraints have to be known and kept by the method.
  • System constraints in an operating theater have to be considered and followed.

Requirements of Different Scenarios

• Cohort study: Very large number of cases and low time constraints.
• Clinical study: Large number of cases and low time constraints.
• Computer-aided diagnosis: Small number of cases and medium time constraints.
• Treatment planning: Small number of cases and medium time constraints.
• Computer-assisted surgery: Small number of cases and high time constraints.

Problem Description

• Defining the problem to be solved and the underlying assumptions for a solution is important.
• Anatomical variation, artefacts from image generation, and changes of appearance due to different imaging techniques are important.
• Problem description comes from discussion between data expert (radiologist, surgeon, physician) and methodology expert (computer scientist, engineer).
• An experienced radiologist has the knowledge to come to a well-founded decision.
• A radiologist is seldom forced to formalize this decision.
• Because of different scientific backgrounds, it will make it difficult for either side to decide on validity, representativeness, and exhaustiveness of any fact.

Required Information for Description Process

• Description of the images used for analysis.
• Description of the patient group.
• All image features used with assumptions about reliability.
• All a priori information used as domain knowledge.

Description and Evaluation

• Test data results in changes in the method to correct errors must result in changes to the description.
• Description helps to set up an evaluation scenario to ensure information reflects underlying truth.
• Evaluation method 1 is evaluated with respect to a proven method and is simple.
• Evaluation method 2 is evaluated with respect to made assumptions a priori.

Data Analysis Tools

• There are different types of software used for data analysis.
• The types of software are viewer software, and analysis software.

Different Software Types

• Viewer software: Primarily developed for viewing data and contains some methods for analyzing the data.
• Analysis software: Provides the user with a set of parameterizable analysis modules for different steps of image analysis.
• Examples of image enhancement methods are segmentation, registration, and classification tools.
• Rapid prototyping software: Use a rapid prototyping programming language such as Matlab or IDL.
• Software libraries: Pays to implement directly into C/C++ or any other suitable language using the techniques and strategies for efficient computation.

Software Libraries

• OpenCV and ITK are mainly used for analysis tasks in medical imaging.
• OpenCV's disadvantage is that it has no 3D or 4D scene processing.
• ITK (Insight Toolkit) focuses on segmentation and registration of medical images.
• Registration methods not included in OpenCV for rigid and nonrigid registration are part of the ITK software.
• Information about OpenCV and ITK can be accessed by their sites.

Web Resources

• A free medical imaging software is “I do Imaging” (http://www.idoimaging.com).

CAD System

• Case 1: A Liver Tumor CAD System includes liver datasets to diagnose malignant/benign tumors. The system incorporates steps for Input, Preprocessing, Enhancing Image Quality, Segmentation, and Classification.
• Case 2: A CAD for Diabetic Retinopathy Grades uses fundus images to classify the images and identify a grade for classification. The system incorporates steps for Pre-Processing and Segmentation steps.
• Case 3: A CAD for Detecting Diabetic Retinopathy Grades uses color fundus images. The system incorporates Processing and Modeling by E-DenseNet steps.
• Case 4: A CAD for Alzheimer's Disease (AD) Grades uses ADNI datasets with MRI and PET scans. The system incorporates Preprocessing Phase, Early Fusion, Late Fusion, and Diagnostic Using XGBoosting steps.
• Case 5: Brain Grades Diagnosis system incorporates Normalization for Intensity correction and ROI Extraction using Extracted Color Features. An additional feature is extraction of textures.

Homework

• Answer the exercises at the end of Chapter 1: pages 18-19.