Artificial intelligence (AI) is redefining the boundaries of medical diagnosis by combining speed, accuracy, and personalization. Since the emergence of the first expert systems in the 1960s, deep learning algorithms have demonstrated their ability to detect anomalies imperceptible to the human eye and optimize patient care. In medical imaging, these advances are helping to detect pathologies, with tools such Allisone the interpretation of dental X-rays. However, the integration of AI in the healthcare field raises major regulatory issues, with the entry into force of the AI Act in Europe and the strengthened FDA guidelines in the United States. In this article, we look back at the evolution of artificial intelligence in diagnostics, its current applications, its regulatory framework, and the innovations it brings to clinical practice.

History of artificial intelligence in medical diagnosis

Origins: the first expert systems (1960-1970)

AI applied to medical diagnosis has its roots in the 1960s with the development of the first expert systems. In 1965, Dendral, designed by Edward Feigenbaum and Joshua Lederberg, was developed to analyze molecular structures using mass spectrometry. Although focused on organic chemistry, it laid the methodological foundations for future medical applications, particularly through its structured approach to problem solving.

A few years later, MYCIN (1972-1976), developed at Stanford University, marked a turning point by specializing in the diagnosis of bacterial infections (meningitis, septicemia). Its inference engine was based on 500 logical rules associated with certainty coefficients (0 to 1), a pragmatic but non-probabilistic method for weighting decisions. Although never deployed in clinical practice, MYCIN achieved 69% acceptable antibiotic prescriptions, a rate comparable to that of human experts.

The rise of probabilistic models and technical limitations (1980–1990)

The 1980s saw the introduction of Bayesian networks to manage uncertainty in medical data. PUFF (1979), an expert system derived from MYCIN, was designed to diagnose obstructive pulmonary diseases by integrating spirometric measurements. Its relative success in hospitals demonstrated the clinical usefulness of rule-based systems.

At the same time, PROSPECTOR (1976-1983), a geological expert system using Bayesian networks, identified a molybdenum deposit in Washington State. Although outside the medical field, this system popularized this method, which was later adopted in certain medical tools.

Artificial neural networks were then the subject of theoretical research for image analysis, but their clinical application remained impossible due to hardware limitations. Systems such as MYCIN, designed for mainframes, required hours of computation and complex text-based interfaces, which slowed their adoption.

The era of big data and deep learning (2000–present)

The 2000s marked a turning point with the explosion of medical data and the advent of deep learning. Convolutional neural networks (CNNs), trained on databases containing millions of images, achieved unprecedented performance levels. According to an article published in 2020, an algorithm developed by Google outperformed several radiologists in identifying signs of breast cancer, reducing false negatives to 9.4% and false positives to 5.7% among thousands of mammograms studied on American patients.

AI extends beyond imaging:

• Hybrid systems combine deep learning and medical knowledge bases to improve differential diagnoses.
• In 2021, models such as CheXNeXt automatically analyze chest X-rays, identifying 14 common conditions with an accuracy of 80 to 94%.

Challenges remain, including the generalization of algorithms to diverse populations and seamless integration into clinical workflows.

What are the fields of application for AI in medical diagnosis?

Computer vision is transforming diagnostic capabilities in medicine. Convolutional neural networks, trained on large databases, identify subtle abnormalities such as breast microcalcifications or vascular lesions. Certain algorithms, such as those developed for chest scans, detect lung nodules as small as 2 mm in diameter, a threshold that is difficult to achieve with the naked eye.

Artificial intelligence is also used in histopathology to automatically sort biopsies, particularly in cases of melanoma. Other tools aim to identify rare diseases by cross-referencing complex clinical data.

In dentistry, it now provides active assistance in the analysis of 2D dental X-rays, serving as a second opinion for diagnosis.

Other areas where AI is playing an increasing role include:

• assessment of surgical risks using intraoperative sensors;
• robotic assistance for implant placement (e.g., da Vinci system);
• assistance with differential diagnosis in general practice;
• early detection of complications through continuous physiological monitoring.

How do regulations govern AI?

Effective July 2024, the AI Act classifies certain medical artificial intelligence systems as high-risk devices, requiring strict compliance with ISO 13485 standards for quality management and IEC 62304 standards for software safety. Manufacturers must ensure the traceability of training data, guarantee the explainability of algorithms, and implement post-market surveillance. Devices are required to document each algorithmic decision and adapt their models to new clinical data.

In the United States, the FDA has accelerated the certification of AI medical devices since 2022, with stricter requirements for ongoing validation, including adaptability to ethnic, geographic, and technological differences, as well as differences between different types of scanners. The FDA is also collaborating with the NIH to develop an open-source database for evaluating and comparing the performance of diagnostic algorithms.

The use of health data is governed by the GDPR and the French Data Protection Act, which require databases to be anonymized and patients to give their informed consent. Despite these safeguards, algorithmic bias remains a major issue. Preliminary work highlights the need to increase the diversity of training data in order to limit ethnic bias and ensure diagnostic fairness.

Allisone : your AI co-pilot for patient communication in dentistry

Artificial intelligence is transforming many areas of healthcare, and dentistry is no exception to this revolution. In dental practices, AI is not limited to automated analysis: above all, it is becoming a tool to support practitioners in streamlining consultations, clarifying explanations, and strengthening patients' understanding of and commitment to their care.

It is precisely in this context that Allisone.

Allisone a digital solution that uses artificial intelligence to highlight elements on a dental X-ray through color coding, educational annotations, and interactive visual aids. The goal is not to diagnose in place of a professional, but to facilitate communication between practitioners and patients around images and proposed treatments.

Our patient survey highlights a major problem in dental communication: 67% of patients are unable to identify conditions on their own X-rays, and 42% feel that their dentist does not share enough information about their images. This lack of visibility and explanation complicates decision-making and hinders adherence to proposed treatments.

Allisone directly Allisone this issue by providing several essential features:

• Color codes to identify the elements visible on the X-ray. 70% of patients believe that such a visual aid would help them better understand their oral health.
• Illustrated care instructions and educational fact sheets enable practitioners to explain each element, thereby avoiding misunderstandings related to technical jargon.
• Post-consultation reports, requested by 70% of patients, providing a written and visual record of the analysis performed, thereby facilitating decision-making.

By making the image easier to read, Allisone understanding, restores a more fluid dialogue in the chair, and allows the patient to engage more calmly in their care journey.

‍

‍

Artificial intelligence has profoundly changed medical diagnosis, evolving from rigid expert systems to tools capable of continuous learning. However, this advance is accompanied by an increased need for regulation to ensure the safety and fairness of practices. The AI Act and ISO standards establish a framework that reinforces the reliability of medical devices. The future lies in hybrid models, where AI assists clinicians without replacing their expertise. With strong annual growth in investment in the European medical AI sector, we are moving towards more predictive and personalized medicine.

In dentistry, tools such as Allisone the relationship between practitioners and their patients. By facilitating the visualization of X-rays and promoting an educational approach, Allisone dentists modernize their practices while improving the patient experience. The goal is clear: to make practitioners' communication more precise, understandable, and interactive.