[An attempt to numerical diagnosis by computed tomography (author's transl)]

Area of Science:

• Radiology
• Medical Imaging
• Biostatistics

Background:

• Previous work established methods for obtaining standard deviation (SD) and deviation coefficient (DC) from brain CT scans of the basal ganglia.
• The potential for numerical discrimination between normal and abnormal scans using these values was previously discussed.

Purpose of the Study:

• To apply admissible linear discriminant analysis for numerical diagnosis of basal ganglia abnormalities using SD and DC values.
• To evaluate the accuracy and reliability of this quantitative method in differentiating normal from abnormal brain CT scans.

Main Methods:

• Linear discriminant analysis was applied to 50 normal and 50 abnormal brain CT scans.
• A discriminant function S = -0.4248(SD) - 6.4709(DC) was derived, with a threshold of S >= -11.447 for normal scans.
• The method's reproducibility was tested on an additional 20 normal and 20 abnormal scans.

Main Results:

• The theoretical misdiscriminant rate was 7.6%.
• The analysis correctly classified 98% of abnormal scans and 92% of normal scans, with an actual misdiscriminant rate of 5.0%.
• In a validation set, only one normal scan was misclassified, resulting in a misdiscriminant rate of 2.5%.

Conclusions:

• The developed method demonstrates high discriminant rates for classifying brain CT scans.
• This quantitative approach is effective for screening examinations due to its high overall accuracy and low false negative rate.