Updated: Aug 26, 1998
This is the most recent structured information on this project. Complementary data can be found on previous documents
CANTOR Converging Agreement by Networking Telematics for Object Recognition

Overview

• In a series of industrial, clinical, and scientific processes we are dependent upon our ability to make visual recognition, i.e. to classify different types of objects. These processes of recognition have one common denominator. They are all partly based on individual evaluation.
• When a medical doctor evaluates the probability of a tissue containing, e.g., cancer cells, these observations are hypothetical evaluations and decisions. They are based on a personal judgment, which at the moment can neither be verified or rejected.
• For the patients it is of utmost importance that doctors and lab technicians possess a high level of visual capability to recognise the objects that they are supposed to judge in the diagnostic phase in accordance with the classification criteria of the clinic or international standards.
• Unfortunately, within many healthcare sectors, considerable inter- and intra-observer variabilities exist. In order to demonstrate the possibilities of reducing these variabilities by use of a new software system and a strategy for Converging Agreement by Networking Telematics for Object Recognition (CANTOR), two healthcare sectors have been selected: histopathology of oncology and serology of autoimmune analyses.
• The software system will be able to perform statistical consensus analysis and to compare agreements and disagreements in object classifications carried out by the same person in repeated processes, by several persons and by groups of persons.
• Within each of the above mentioned sectors, CANTOR aims at demonstrating the use of telematics for three important stages related to visual classification of objects in microscopic images: standardisation, quality assurance and education.
• Within each healthcare sectors, reference images will be digitised by video-microscopy and then distributed on CD-ROM. In combination with continuously consensus improving visual knowledge databases, transfer of visual knowledge via Internet will be demonstrated as to attain converging agreement in object classification.
• The procedure within CANTOR will be to elicit detailed user requirements for updating existing software with respect to the application of such a system for standardardisation, quality assurance and education in object classification. Based on converging agreement in the visual knowledge databases, reduction of the intra- and inter-observer variability will be demonstrated.

Purpose and objectives

1. Further develop DOORS (Discrete Object Observation and Recognition System) such that it becomes a tool that can be used for performing an analysis of visual interpretation and classification of (objects in) images in three different settings: Education and Training, Quality Assurance and Standardisation.
2. To perform a validation of this extended DOORS in the three different settings in two medical domains: Pathology/oncology and auto-immune diseases.

More detailed we will further develop DOORS into a tool that will
In an educational setting:
Provide a means for doing a performance reproducibility test on the visual interpretation of the same images.
Provide a cost effective means for training in the classification of objects in an image, e.g. cells which are marked with a certain tumour marker.
Provide direct graphical feedback to the student on which objects are properly classified and which are not.
Provide a means for doing a test as DOORS allows for comparison between some reference classification and the users classification.
Focus discussions with the teachers to the objects for which it is unclear why classifications are not according to the reference.
Log any changes in performance during a training course.
In an Quality Assurance setting:
Allows various members of a service to analyse the same images. Provide measures on the individual performance (either compared against a reference or against other members of the service).
Allows to analyse the intra- and interobserver as well as intergroup variability.
Focuses on areas of improvement (intra-observer variability, poor agreement with colleagues etc.).
In a consensus formation setting: Provides support to measure at various levels the variability on classifications: intra-observer, between observers in one unit or organisation, between groups of observers.
Provides various modes of presentation of results as to give insight on problematic issues that need further discussion.
Allow involved specialists to participate from their own office as results will be stored and made available through WEB technology.

The proposed tool will be of value for any medical discipline in which the interpretation of objects in two-dimensional images is relevant. As a practical example we envision that the extended version of DOORS could be used as a tool for certifying organisations for e.g. breast or cervix cancer screening as well as means to assist in keeping the quality of such services at the required level.

Results

A result from the consensus formation process is that a set of reference images becomes available, together with a reference classification for each of the relevant objects in those images. Such reference material could be made available on CD-ROM or the Internet by e.g. publishers.

The consensus committees of various professional organisations could utilise DOORS in their process. Running DOORS connected with a WEB server would make it feasible to distribute images and collect visual classifications while the participants are in their own environment. The analysis software of DOORS would then indicate areas of disagreement and the consensus formation process could be speeded up. It is expected that this type of utilisation of DOORS would significantly reduce the cost associated with consensus formation.

CANTOR plans to attend several conferences and exhibitions of professional societies as to present the extended DOORS and to demonstrate its value. Awareness of the target users will increase the possibilities for successful commercial exploitation of the results of CANTOR.

List of deliverables

Year 1

• D01.1 Project Handbook (02, R)
• D10.1 Quality Manual.( 02, R)
• D01.2 Success Indicators CANTOR (03, L)
• D02.1 External Communication Plan (03, R)
• D05.1 Quality Assurance Procedures for Image Data Base (05, P)
• D03.1 User Requirements for upgrading the existing software with respect to education and training, quality assurance and consensus formation. (05, R)
• D05.2 Assessment Plan (06, L)
• D04.1 Prototype Extended DOORS for Verification (10, R)
• D05.3 Image Data Bases (12, P)

Year 2

• D06.1 Educational Material (16, L)
• D07.1 Quality Assurance Protocol (16,L)
• D08.1 Protocol for Consensus Formation Process (16,L)
• D04.2 Extended DOORS (18,L)
• D06.2 User Acceptability Report Education and Training (18,R)
• D07.2 User Acceptability Report Quality Assurance (18,R)
• D08.2 User Acceptability Report Standardisation (18,R)
• D09.1 On the Usability of Extended DOORS for Education and Training, Quality Assurance and Consensus Formation (23, P)
• D01.3 Exploitation Plan (24,L)
• D02.2 CD-ROM master of Reference Image libraries (24,L)
• D10.2 Project Assessment (24, R)

List of participants

	Copyright 1998 © EHTO All rights reserved This server is the only official EHTO WWW knowledge repository. Mail suggestions to: webmaster@ehto.org