Computer-assisted detection

from Wikipedia, the free encyclopedia

Computer-aided diagnosis ( computer-assisted detection , partly computer-aided diagnosis , short CAD ) describes a method in medicine to assist the clinician in the interpretation of test results.

Imaging methods in X-ray diagnostics provide a wealth of information that the radiologist must comprehensively analyze and evaluate in a short time. CAD systems help to search digital image data, for example from computed tomography , for typical patterns and to visually highlight conspicuous areas (possible diseases).

CAD is a relatively young interdisciplinary technology and combines elements from artificial intelligence and digital image processing with radiological image processing. A typical area of ​​application is tumor diagnostics. CAD supports preventive mammography examinations (breast cancer diagnostics), the detection of polyps in the colon and lung cancer .

Use

CAD systems are usually limited to marking conspicuous structures and areas. In addition, computer-assisted diagnosis systems (computer-assisted diagnosis - CADx) are used. The system also evaluates the conspicuous structures.

For example, CAD in mammography marks changes in soft tissue or micro-calcification in X-ray images in different ways. This leads to further conclusions about the nature of the pathology. Another form is CADq for quantifying z. B. of tumor size or the contrast agent uptake behavior of tumors.

At present, CAD cannot and must not replace the medical professional and is only of supportive importance. In any case, the final assessment and responsibility for the diagnosis made rests with the individual doctor.

application areas

In addition to the diagnosis of breast and lung cancer, further areas of application are the detection of colon cancer and prostate cancer .

Breast cancer (breast cancer)

The main area of ​​application is in mammography (X-ray examination of the female breast). In the form of a screening , mammography has been used for many years for the preventive early diagnosis of breast cancer . CAD is mainly established in the USA and the Netherlands and serves the diagnostician as a second opinion for human evaluation. The first CAD system for mammography was developed as part of a research project at the University of Chicago. It is offered commercially today by the company R2. There are also methods for evaluating MRT-based mammography ( magnetic resonance tomography ).

Lung cancer (bronchial carcinoma)

Computed tomography with special 3-dimensional CAD systems has established itself as the gold standard in lung cancer diagnostics. Here, a volumetric data set from up to 3,000 individual images is processed and analyzed. Round nodules (lung cancer, metastases and benign changes) from 1 mm can be detected. All major manufacturers of medical systems now offer corresponding solutions.

Sensitivity and specificity

CAD systems should reliably mark conspicuous structures. However, today's CAD systems cannot 100% recognize pathological changes. The hit rate ( sensitivity ) is up to 90% depending on the system and application.

A correct hit is called a true positive (TP). At the same time, healthy areas are also marked, which are referred to as false positive (FP). The fewer FPs displayed, the higher the specificity . Too low a specificity reduces the acceptance of a CAD system, since these false hits have to be identified individually by the radiologist each time. The FP rate on lung overview images (CAD Chest) has already been reduced to approx. 2 per examination. In other areas (e.g. CT lung exams) it can be 25 and more.

Absolute recognition rate

More important than sensitivity and specificity is the radiologist's absolute detection rate. Depending on experience, training and application, CAD systems can help to increase the recognition rate. In mammography, the increase is on average 20-30%. The early detection of pulmonary nodules can be increased by more than 50%.

In general, study results on sensitivity, specificity and the absolute detection rate can vary widely. The results depend on the given framework and must be evaluated on a case-by-case basis. The following factors have a major influence:

  • Retrospective or prospective study design
  • Quality of the image material used
  • Conditions for taking the X-ray images
  • Experience and training of the observer / radiologist
  • Type of disease / tumor
  • considered tumor size

methodology

CAD is essentially based on highly complex pattern recognition . X-rays are searched for abnormal structures. As a rule, a few thousand images are required to optimize the algorithm. Digital image data are transferred to a CAD server in DICOM format and processed and analyzed in several steps.

1. Preprocessing for

  • Reduction of artifacts (image defects)
  • Reduction of the picture noise
  • Leveling the image quality to accommodate the different conditions under which the image was created, e.g. B. different recording parameters.

2. Segmentation for

  • Delimitation of the different structures within the image, e.g. B. Heart, lungs, ribs, possible round nodules
  • Comparison with anatomical databases

3 . Structure / ROI (Region of Interest) analysis

Each recognized region is individually analyzed for special characteristics. These are u. A.

  • compactness
  • Shape, size and location
  • Relation to neighboring structures / ROIs
  • Average gray value distribution within the ROI
  • Ratio of the gray values ​​within the ROI to the edge of the structure

4. Assessment / classification

After the structural analysis, each ROI is individually assessed ( scoring ) in order to determine the probability of a really positive hit. Procedures for this are:

  • Artificial Neural Network ( ANN )
  • minimum distance classifier
  • Cascade Classifier
  • Bayesian filter
  • Multilayer perception
  • Radial basis function network (RBF)
  • SVM

If the structures found have reached a certain threshold value, they are marked in the image for the radiologist. Depending on the CAD system, all markings are permanently documented (saved) or only temporarily. The latter has the advantage that only the markings confirmed by the radiologist are saved. Incorrect hits should not be documented, as this can make it more difficult to view the image data later.

Individual evidence

  1. ^ T. Wollenweber, B. Janke, A. Teichmann, M. Freund: Correlation between histological findings and a computer-assisted detection system (CAD) for mammography. Obstetrics Frauenheilk 2007; 67: 135-141 doi : 10.1055 / s-2006-955983 .