Computer-aided diagnosis

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

Imaging techniques in X-ray diagnostics provide a wealth of information that must be comprehensively analyzed and evaluated by radiologists in a short time. CAD systems help digital image data, for example, computed tomography scan according to typical patterns and problem areas (possible diseases) visually emphasize.

CAD is a relatively young interdisciplinary technology and combines elements of artificial intelligence, and digital image processing, with radiological imaging. Typical application is tumor diagnosis. It supports the CAD screening mammography ( breast cancer diagnosis ), the detection of polyps in the colon and lung cancer.

Benefit

CAD systems are limited usually sure to highlight conspicuous structures and areas. In addition to this computer-assisted diagnosis systems (computer -assisted diagnosis - CADx ) are used. Here, in addition an assessment of eye-catching structures is performed by the system.

CAD in mammography labeled, for example soft tissue change or microcalcification in X-ray images in different ways. This results in further inferences about the nature of the pathology result. Another form is CADq to quantify, for example, tumor size or the contrast medium uptake behavior of tumors.

Currently, can not and must not replace the physician CAD ​​and has only supportive importance. In any case, makes the final assessment and the responsibility for the diagnosis determined the individual physician.

Areas of application

In addition to the diagnosis of breast and lung cancer other applications are the detection of colon cancer and prostate cancer.

Breast cancer ( breast cancer)

The main field of application is in mammography ( X-ray examination of the breast ). In the form of routine screening (screening) mammography is used for preventive screening for breast cancer for years. CAD is here especially in the U.S. and the Netherlands established and serves the diagnostician as a second opinion for human evaluation. As part of a research project at the University of Chicago, the first CAD system has been developed for mammography. It is commercially available from the company R2 today. Methods also exist for the evaluation of MRI -based mammography (magnetic resonance imaging ).

Lung cancer ( lung cancer )

The gold standard in lung cancer diagnosis, computed tomography with special 3-dimensional CAD systems has been established. In this case, a volumetric dataset consisting of up to 3,000 individual images is processed and analyzed. Nodules (lung cancer, metastases and benign changes ) from 1 mm can be detected. All major manufacturers of medical systems today offer appropriate solutions.

Sensitivity and Specificity

CAD systems are intended to highlight conspicuous structures reliably. Nevertheless, today's CAD systems pathological changes not seen at 100%. The hit rate (sensitivity ) is depending on the system and application at up to 90%.

A real hit is called True Positive ( True Positive, TP ). At the same time healthy areas are marked as false positive (false positives, FP) are indicated. The less FP are displayed, the higher the specificity. Too low specificity reduces the acceptance of a CAD system, since these false hits by the radiologist every time need to be individually identified. The FP rate on lung overview images (CAD Chest ) has already been reduced to about 2 per examination. In other areas (eg CT lung studies ) it may be 25 or more.

Absolute detection rate

More important than sensitivity and specificity is the absolute recognition rate of the radiologists. Depending on experience, training and application CAD ​​systems can help to increase the recognition rate. In mammography, the increase in average 20-30%. Early detection of pulmonary nodules can be raised by more than 50 %.

In general, study results for sensitivity, specificity, and the absolute recognition rate can vary widely. The results are dependent on each of the given conditions, and must be evaluated on an individual basis. The following factors have a major influence:

  • Retrospective or prospective study design
  • Quality of the photographic material used
  • Recording condition of the radiographs
  • Experience and training of the observer / radiologist
  • Type of disease / tumor
  • Considered tumor size

Methodology

CAD is mainly based on highly complex pattern recognition. X-ray images are searched for eye-catching structures. Usually a few thousand images are required to optimize the algorithm. Digital image data is transferred in DICOM format to a CAD server and processed in several steps and analyzed.

First preprocessing for

  • Reduction of artifacts ( artifacts )
  • Reduction of the image noise
  • Leveling the image quality to meet the different conditions under which the image was created to compensate, for example, the recording parameters.

Second segmentation for

  • Delineation of the different structures within the image, such as heart, lungs, ribs, possible nodules
  • Matching with anatomical databases

3 structure / ROI ( region of interest) analysis

Each detected region is analyzed individually to specific characteristics. These are u A.

  • Compactness
  • Shape, size and location
  • Relation to adjacent structures / ROIs
  • Average gray level distribution within the ROI
  • Ratio of the gray values ​​within the ROI to the edge of the structure

4 Rating / Classification

According to the structure analysis, each ROI is individually assessed ( Scoring ) thus to determine the probability of a true positive results. Methods for this are:

  • Artificial neural network (ANN )
  • Minimum distance classifier
  • Cascade Classifier
  • Bayesian filter
  • Multilayer perception
  • Radial basis function network ( RBF)
  • SVM

Do the structures found reaches a certain threshold, they will be marked in the image for the radiologist. Depending on the CAD system, all markings are permanently documented (recorded), or only temporary. The latter has the advantage that only the flags are stored confirmed by the radiologist. Wrong results should not be documented because it may complicate subsequent analysis of the image data.

Medical imaging
199430
de