Search

This paper describes the initial findings of computer-assisted diagnosis of cerebral tumours using Bayes' Theorem and CT scans. For various sets of signs, tables of accuracy and reliability have been constructed, and a computer versus radiologist analysis for three types of tumour has been made. Based only on the results of CT scans, the computer has produced diagnoses relatively quickly, and overall these are of reasonable accuracy when compared with the radiologists' reports. The worst results occurred in the prediction of meningiomas, the reasons for which will be examined in subsequent work.

High-resolution computerized tomography (CT) has been used to investigate abnormalities of the ear and petrous bone. Congenital, traumatic, infective and neoplastic lesions have been evaluated in 30 patients with proven pathology. Some bone changes including abnormalities of the auditory ossicles have been better demonstrated than by conventional tomographic technigues, and high-resolution CT has some advantages in demonstrating both soft tissue abnormality and bone changes in the middle ear in primary neoplastic disease and secondary cholesteatoma. The absence of any practical means of taking sagittal sections with the present scanner design limits the usefulness of high resolution CT of the petrous bone, but this study has shown that the technique may now take its place as an important complementary procedure to pluridirectional tomography.

Eighty colour subtractions of cerebral angiograms have been analysed to assess their value in actual diagnosis, particularly of cerebrovascular disease. The method has been found to be so valuable, either for confirmation of the findings seen on the original angiograms or in providing new data about the accurate localization and nature of the lesion, that it has now become a routine for all cerebrovascular cases. A selection of cases is presented and the limitations of the method discussed.

Endocrinologists tend to accept minor variations of the pituitary fossa as evidence of the presence of a pituitary gland abnormality. Neuroradiologists tend to be sceptical. Normal criteria have not been established. In a consecutive series of 85 cases without any clinical evidence of a pituitary lesion, the plain films in 31.7 per cent demonstrated minor duplication of the floor or anterior wall of the pituitary fossa (with or without associated thinning and possible erosion of the lamina dura), which could be mistaken for the early signs of pituitary enlargement by an intrasellar tumour. 16.5 percent showed thinning of the lamina dura on tomography a feature previously taken as suggesting early enlargement. It is suggested that these particular diagnostic signs should be interpreted with caution.

Regional cerebral blood volume (rCBV) was measured in 14 patients with normal CT scans. An EMI CT 1010 scanner was used in combination with a computer subtraction technique. The mean CBV in the cortex was 5.0 ml/100 ml of tissue and 2.2 in the white matter. Regional differences were not significant and no difference was found between the right and left hemispheres.

A computerized subtraction technique has been described to measure regional cerebral blood volume (rCBV) using the EMI-Scanner in a group of 13 patients. Sodium iothalamte was injected intravenously (1-75 ml./kg) to increase the absorption of X rays in the cerebral circulation. Significant regional differences in CBV were shown, values in the frontal and temporal regions being lower than the mean hemisphere value (4-9 +/- 0-7) and higher in the occipital region. The left hemisphere showed a significantly higher CBV when compared with the right. Measurements of CBV in the cortex showed no regional variation, but the mean cortical value of 6-0 +/- 1-8 was significantly higher than the hemisphere mean.

Most applications of Bayes theorem in computer-aided diagnosis have been to situations involving the differential diagnosis of a small list of disease categories. In the case of the diagnosis of cerebral tumours, if each tumour type in each main anatomical situation is counted as a single diagnosis, there are about 100 possible disease categories. This paper investigates a method of incorporating expert prior information into the computer-aided diagnosis process so that this large number of categories can be handled.