What Is Computer Aided Diagnosis?

Computer aided diagnosis (CAD) is the use of computational methods to analyze medical data, primarily medical images, and generate quantitative assessments that assist clinicians in detecting disease, characterizing lesions, or reaching diagnostic conclusions. A CAD system functions as a second reader: it processes imaging studies through automated algorithms and highlights findings or provides probability scores that the physician reviews alongside the raw data.

The concept emerged in academic radiology in the 1980s, with early systems targeting the detection of microcalcifications in mammography. Since then, the scope has expanded to cover virtually every imaging modality, including computed tomography, magnetic resonance imaging, ultrasound, digital pathology, and nuclear medicine, as well as non-imaging data streams such as electrocardiograms, genomic profiles, and electronic health records. CAD now divides into two functional categories: computer aided detection (CADe), which locates potential abnormalities for the clinician to evaluate, and computer aided diagnosis proper (CADx), which characterizes detected findings and estimates likelihood of malignancy or disease.

Image Processing and Feature Extraction

Classical CAD pipelines begin with image preprocessing steps such as noise reduction, contrast enhancement, and organ segmentation, followed by the extraction of quantitative features from candidate regions. Morphological features describe shape and margin characteristics; texture features capture local intensity patterns; kinetic features summarize how tissue enhances over time after contrast injection. These handcrafted features feed into statistical classifiers, support vector machines, or shallow neural networks trained on annotated datasets to produce diagnostic scores. The design of informative, reproducible features was the central engineering challenge of first-generation CAD, and feature selection methods including principal component analysis and recursive feature elimination were essential to managing high-dimensional feature spaces. A broad survey of this era and its transition to deep learning is available in a PMC review of CAD in the era of deep learning.

Deep Learning Architectures

The adoption of convolutional neural networks (CNNs) from the mid-2010s onward changed CAD fundamentally. Rather than requiring expert-designed features, CNNs learn hierarchical representations directly from pixel data, with early layers detecting edges and textures and deeper layers encoding complex anatomical patterns. Architectures such as U-Net, which was designed for biomedical image segmentation, and detection frameworks derived from Faster R-CNN and YOLO have been adapted for radiology and pathology tasks. Deep learning-based CAD applications reviewed in Sensors document performance across modalities including CT lung nodule detection, mammographic mass characterization, and retinal fundus analysis. Transformer-based models trained on large multi-center datasets have more recently extended CAD toward tasks that require integrating image content with clinical context, such as report generation and structured finding extraction.

Validation and Clinical Integration

Deploying a CAD system in a clinical setting requires evidence that it improves diagnostic outcomes without introducing unacceptable error rates. Regulatory pathways in the United States run through the FDA's Software as a Medical Device framework, which requires analytical and clinical validation studies. Observer performance studies using receiver operating characteristic (ROC) analysis are the standard for measuring how much a CAD system changes clinician sensitivity and specificity. Large retrospective datasets and prospective reader studies have demonstrated measurable improvements in cancer detection rates for specific tasks, while highlighting that CAD performance can degrade with scanner variation, patient population shift, and imaging protocol changes. The University of Chicago radiology program on CAD and machine learning has contributed foundational methodology to this validation literature.

Applications

Computer aided diagnosis has applications across a range of medical imaging and clinical data domains, including:

• Pulmonary nodule detection and lung cancer screening in CT
• Mammographic lesion characterization and breast cancer screening
• Diabetic retinopathy screening from fundus photographs
• Cardiac function quantification from echocardiography and MRI
• Histopathology slide analysis for cancer grading and staging
• Neurological assessment from brain MRI in dementia and multiple sclerosis