FFCLRP helps dentists read x-rays, speeding up diagnosis

A study published in the journal PLOS Digital Health demonstrates the accuracy of an artificial intelligence system developed at USP in Ribeirão Preto for identifying teeth and detecting cavities in x-rays. Approximately 30 researchers from the Faculty of Philosophy, Sciences and Languages at Ribeirão Preto and the School of Dentistry of Ribeirão Preto are collaborating within the Interdisciplinary Research Group in Digital Dentistry (InReDD) on the project.

According to professor Alessandra Alaniz Macedo, the technology is based on Convolutional Neural Networks (CNNs), a type of artificial intelligence specialized in image processing; she explained that the system aims to streamline diagnosis and treatment by assisting dentists in radiographic analysis.

Convolutional Neural Networks Detect Cavities in Dental X-rays

The University of São Paulo’s Ribeirão Preto campus has developed an artificial intelligence system capable of identifying cavities in dental x-rays with a high degree of accuracy, streamlining diagnostic workflows for dentists. This capability stems from the implementation of Convolutional Neural Networks, or CNNs, a specific type of artificial intelligence architecture designed for image processing and pattern recognition.

According to professor Alessandra Alaniz Macedo, “Within artificial intelligence, different methods are used to analyze different types of data. Neural networks are one of these approaches and currently produce the best results for various types of problems.” Training the AI to reliably detect cavities requires a supervised learning process where the system is fed thousands of previously analyzed radiographs. Researchers create a dataset, essentially teaching the artificial intelligence to correlate specific image characteristics with the presence of lesions.

The model then iteratively compares its own analysis of new radiographs against this established ground truth, correcting errors and refining its accuracy with each iteration. Researchers explained that the model continuously compares its output with the ground truth and corrects its own errors during training, gradually learning to identify lesions with increasing accuracy.

This data preparation and validation is primarily the work of computer scientists specializing in artificial intelligence. Beyond cavity detection, the InReDD team is exploring how AI can automate administrative tasks, schedule appointments, and even assist in treatment planning within dental practices, the company says. Despite the potential for increased efficiency and diagnostic support, the researchers emphasize that the technology is intended to augment, not replace, the expertise of dentists.

Professor Camila Tirapelli noted that current AI models operate as a “black box,” delivering accurate answers without always revealing the reasoning behind them, and that all AI systems are susceptible to errors. “Responsibility for the diagnosis will always lie with the professional. No model is perfect, and we know they will always make mistakes, even if only rarely,” she said.

The quality of the AI’s output is directly dependent on the quality of the training data; if humans teach it incorrectly, AI will learn incorrectly. That is why healthcare and computing professionals must work together to develop accurate models.

The system’s code is registered by USP, and the University views it as a Brazilian initiative integrating multiple functions typically found in separate products.

AI as Support Tool: Accuracy, Limitations, and Future Research

This group focuses on translating algorithms into practical tools for dental professionals and researchers, evidenced by the system’s registered code developed within USP’s laboratories. These networks aren’t simply identifying shapes; they’re learning to recognize patterns indicative of dental lesions, a process that involves continuous self-correction during training.

This collaborative approach extends beyond model development, encompassing data preparation and validation, all areas of expertise within the InReDD team, which includes undergraduate, master’s, and doctoral students alongside faculty members and postdoctoral researchers. This opacity, coupled with the inherent possibility of errors, underscores the importance of maintaining human oversight.

More like this

Artificial IntelligenceRats Beat AI in Visual Recognition Tests

Artificial IntelligenceWiMi’s Quantum Circuits Boost Image Recognition Efficiency

Technology NewsYann LeCun: AI & Convolutional Neural Networks

TechnologyOptical CNN Achieves Image ID with 24 Kernels