Use this URL to cite or link to this record in EThOS:
Title: Applications of signal detection theory to the performance of imaging systems, human observers and artificial intelligence in radiography
Author: Manning, David J.
ISNI:       0000 0004 2736 4170
Awarding Body: Lancaster University
Current Institution: Lancaster University
Date of Award: 1998
Availability of Full Text:
Access from EThOS:
Access from Institution:
An investigation was carried out to evaluate diagnostic performance in medical radiology. A critical review of methods available for the assessment of image quality in terms of physical objective measurements and quantitative observer performance was followed by a series of experiments which applied the techniques of Receiver Operating Characteristics (ROC) to radiographic problems. An appraisal of the performance of six currently available imaging systems designed for chest radiography was performed using expert observers and an anthropomorphic phantom. Results showed a range of abilities to demonstrate pulmonary nodules (ROC areas of 0.866 to 0.961). The ROC outcomes for the imaging systems were shown to correlate well with signal to noise ratio (SNR) measurements for the images (0.78, p< 0.05) although comparisons of ROC and threshold detection indices (HT) gave a lower level of agreement (0.6, p<0.05). The SNR method of image evaluation could probably be used as an alternative to ROC techniques in routine quality assurance procedures when time is short. Observers from a group of undergraduate radiography students were tested by an ROC study into their ability to detect pulmonary lesions in chest images. Their ROC areas (Az) ranged from 0.616 to 0.857 (mean 0.74) compared with an expert mean score of 0.872. The low score for the students was investigated in terms of the cognitive task and their search strategy training. Their (Az ) scores showed no significant correlation with simple psychometric tests. A neural network was tested against radiologists, radiographers and student radiographers in its ability to identify fractures in wrist radiographs. All observers performed to a similar level of ROC Az score but the artificial intelligence showed higher specificity values. This attribute was used to filter some of the normals from the test population and resulted in changes to the mean Az score for the human observers.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available
Keywords: QA75 Electronic computers. Computer science