TechTarget.com/healthtechanalytics

https://www.techtarget.com/healthtechanalytics/definition/computer-aided-diagnosis

What is computer-aided diagnosis?

By Stephen J. Bigelow

Computer-aided diagnosis is the use of sophisticated machine learning and deep learning techniques to help healthcare providers recognize and diagnose medical conditions in medical image data.

Computer-aided diagnosis (CAD) platforms are commonly used to assist in the diagnosis of image-dependent testing -- such as heart disease and defects, cancer, major organ function, circulatory function, and some brain conditions like Alzheimer's disease.

Medical imaging has been a fundamental tool for medical diagnostics since 1896, when X-rays were first discovered and used to expose photographic plates. X-ray images allow physicians to see inside patients, to detect and identify injury, disease and other physical anomalies. Since then, other imaging technologies have been developed and deployed -- such as magnetic resonance imaging (MRI), computed tomography (CT) scans, endoscopy and ultrasound. These are all intended to provide useful and objective diagnostic data for physicians.

However, medical images contain a tremendous amount of information. Healthcare providers are under pressure to evaluate this imaging data quickly and accurately while not missing any points of concern.

CAD systems use computers to process images through trained machine learning (ML) models. This allows the CAD platform to recognize typical physiology and identify atypical points in the images. This capability provides two key benefits to medical practitioners, including the following:

• Verifying the practitioner's diagnosis, raising confidence in the diagnosis and subsequent treatments.
• Identifying potential abnormalities the practitioner may have overlooked or discounted, allowing a more thorough and accurate diagnosis, and potentially altering the appropriate treatment plan.

Computer-aided diagnosis is sometimes denoted as CADx to distinguish it from similar terminology, such as computer-aided detection -- dubbed CADe. These notations also prevent confusion with the traditional acronym for computer-aided design (CAD). Moreover, CADx and CADe technologies are intended as supporting tools rather than direct replacements for experienced physicians, but emerging ML/AI platforms are systematically making autonomous diagnoses with less oversight from medical practitioners.

How does computer-aided diagnosis work?

CAD technologies use a mix of image processing, machine learning, and AI to ingest, analyze, recognize and identify atypia within medical images. The CAD process has as many as five broad phases.

Prepare images

Images are captured, stored and preprocessed. Preprocessing fundamentally cleans up the raw image, removes artifacts and noise, adjusts image characteristics such as brightness and contrast, and applies any necessary filtering that allows the prepared image to be ingested and analyzed.

Analyze images

Images are then ingested into the CAD system for analysis. CAD systems are based on ML models and algorithms that are previously trained on vast data sets built with thousands of images. This training data routinely includes countless examples of physiological features, atypia, injury and disease.

The ingested images are analyzed against this data set, allowing any deviations or suspicious points to be located in the images' data patterns. As one example, a CAD system trained to detect breast cancer can look for unusual masses and signs of tissue calcification, which might provide early indications of the disease.

Detect regions of interest

Analysis of the ingested images may spot specific regions of interest (ROIs) where possible atypia may be present. These ROIs are not diagnoses but instead serve as general highlights prompting closer attention from radiologists and other medical professionals. As an example, CAD systems trained to process chest radiographs might spot dense areas in the lungs, which might suggest lung nodules -- small abnormal growths in the lungs that might or might not be cancerous.

Provide quantitative assessments

The CAD system can usually provide additional details and metrics. For example, it could estimate the patient's age by measuring the characteristics of joint tissue or report the overall size and shape of an anomaly such as a mass or lesion.

The CAD platform can also offer suggestions for possible diagnoses based on the identified features. For example, a mass of a certain size, shape and density in a patient of a particular age might include a probable diagnosis intended to help radiologists make a more detailed assessment. It can also improve decision-making for further actions -- such as additional testing or more invasive procedures like biopsies for a definitive diagnosis.

Subsequent learning

Once a definitive diagnosis is established and a treatment plan is implemented, the analyzed images can be added to the model's data set, further training the ML model and AI system with image data and determined outcomes. This ongoing training will refine the CAD system and enhance its detail and accuracy in the future.

Benefits and drawbacks of computer-aided diagnosis

The use of computers in medical diagnoses is certainly not new, but the emergence of ML/AI has dramatically enhanced the speed and capabilities of CAD technologies. This brings an array of pros and cons.

Some benefits include the following:

• Earlier diagnoses. CAD systems can locate and potentially identify ROIs that are small enough for radiologists to miss or discount. This can lead to earlier diagnoses and fewer missed diagnoses.
• More accurate diagnoses. CAD systems can enhance the accuracy of image diagnoses, reducing the potential for false positives and yielding more reliable diagnoses. This leads to more successful treatments and better patient outcomes.
• Faster diagnoses. CAD systems can yield results faster than collaborative reviews by working groups of human medical providers. Faster diagnoses can allow earlier treatments and improve the chances for successful patient outcomes.
• Reduced bias and broader knowledge. CAD systems can be trained on vast data sets with perfect recall, rendering objective and consistent diagnostic results without the subjective interpretations that often accompany human review.
• Cost savings. Although CAD systems can be expensive, their service can reduce medical costs through easier, earlier, less-invasive or unnecessary procedures, which can save money for medical organizations, insurers and patients.

Despite the benefits, CAD systems present numerous potential drawbacks that should be considered when adopting and using CAD technologies. Common drawbacks include the following:

• Data dependency. CAD systems rely on ML models and algorithms that require comprehensive training with vast data sets. Data must be plentiful and of high quality to ensure analyses are as accurate as possible. Poor or misrepresented data can result in inaccurate results that can diminish patient outcomes.
• No standardization. There is no set of uniform standards established to guide the development, training, deployment or use of CAD systems. Each system can be different in its design, training data and user interactions. This can lead to poor diagnoses, low performance and inefficient use.
• Incorrect diagnoses. CAD systems are not perfect. No matter how much high-quality data is ingested and regardless of the sophistication of ML models and algorithms, CAD systems can still yield false positives and false negatives. CAD systems are not a replacement for human oversight and medical expertise.
• Complex and costly. CAD systems can be expensive to acquire, deploy, train, maintain and integrate into existing medical workflows. They demand constant attention and training -- all of which can make CAD costly in terms of time, money and technical expertise.
• Ethical issues. CAD systems have enormous potential to aid clinicians, but this can lead to over-reliance on CAD output, where clinicians ignore their own judgment and expertise in favor of the CAD output. There is further ethical concern over the ways that data is selected and analyzed, as well as patient privacy issues related to the use of their image data and personal health information.

Top use cases for computer-aided diagnosis

CAD systems are widely adopted in a range of medical diagnostic use cases, including the following:

• Structural problems. CAD systems can identify failures in the bone structure -- such as improper bone formation or relationships, breaks and fractures, and potential problems with ligaments and other musculoskeletal tissues.
• Cancer detection. CAD systems are adept at detecting and identifying cancerous masses, including subtle precursor signs that can indicate early onset of cancers -- such as the microcalcifications that can suggest the emergence of breast cancer. Similarly, CAD can analyze CT scans to identify lung nodules -- which can signify signs of lung cancer -- as well as bone metastases that indicate cancer that has spread to the bones.
• Brain disorders. CAD systems can analyze varied brain imagery -- such as X-rays, CT and MRI scans -- for indications of common neurological conditions like stroke, aneurysms and Alzheimer's disease.
• Heart and circulatory disorders. CAD systems can review cardiac imaging and analyze heart monitoring data -- such as EKGs and ECGs -- to detect heart problems like arrhythmias, coronary artery disease and heart failure.
• Vision disorders. CAD systems can analyze images of the eye's retina to identify common vision issues such as diabetic retinopathy and other damage to the retina.

History of computer-aided diagnosis technology

Computers have been in medicine almost as long as the computer itself. Modern CAD systems represent advancements in computer-based medical diagnoses, which can be traced back to almost the dawn of the computer age. Major milestones in the evolution of CAD include the following:

• 1960s (expert systems). The use of computers in medical diagnoses started as early as the late 1950s, but offered little intelligence and no modeling. Instead, most "expert systems" relied on flowcharts, knowledge bases and statistical methodologies to assist clinicians in their decisions.
• 1980s (diagnostic assistants). Advancements in computer technologies -- such as the personal computer, significant storage and computer networks for exchanging data -- enabled the creation of fundamental diagnostic aids. The Kurt Rossman Laboratories at the University of Chicago are credited with the early development of CAD systems for the detection of microcalcifications on mammograms.
• 1990s (CAD emergence). CAD methodologies and technologies continued to improve -- allowing studies such as quantitative image analysis and tumor heterogeneity -- further enhancing the scope and capabilities of CAD systems. The USFDA approved CAD systems to assist breast cancer screenings in 1998.
• Current. Ongoing improvements in medical imaging, advances in computer processing and storage, and the emergence of ML have merged to allow CAD systems to detect and identify atypia that are often overlooked by human clinicians. This furthers CAD systems as viable diagnostic tools.

Future of computer-aided diagnosis technology

Advancements in ML/AI technologies have the potential to develop and grow healthcare CAD systems. These advances will make CAD systems more sophisticated and reliable, improving accuracy and healthcare efficiency and opening CAD to other medical areas such as pathology and personalized patient care.

At the same time, CAD systems can handle many routine tasks such as analysis and report generation, letting clinicians focus on patient interactions and more complex patient care issues. Finally, expect CAD systems to settle on some standardizations -- such as data formats -- and provide integrations with other healthcare technologies, such as electronic health records.

However, the speed and ease of these future advancements will depend largely on addressing CAD challenges such as data privacy, bias management and other ethical concerns.