AI for prostate cancer diagnosis is being explored as a way to help urologists interpret ultrasound images and identify areas that may warrant biopsy. The approach combines higher-resolution imaging with machine learning analysis, with the aim of supporting more informed decisions about where to sample tissue.
It addresses a challenge in prostate cancer assessment: conventional ultrasound can guide needle biopsies, but its image resolution may not be sufficient to distinguish suspicious regions reliably. The technology described here pairs a specialised ultrasound system with artificial intelligence trained to analyse its images.
Why prostate ultrasound imaging can be difficult to interpret
Ultrasound is used to guide prostatic needle biopsies, which help clinicians investigate possible prostate cancer. However, the resolution of conventional systems can make it difficult to differentiate suspicious tissue from surrounding areas. The whitepaper reports a false-negative rate of 30 per cent for the current standard-of-care ultrasound approach.
A false negative means a test does not identify a condition that is present. In the context of biopsy guidance, difficulty distinguishing regions of concern can make it harder to target the areas most likely to provide useful diagnostic information. The figures describe the challenge discussed in the source material; they should not be interpreted as a performance measure for the newer system.
High-resolution ultrasound provides a closer view
ExactVu, developed by Toronto-based Exact Imaging, operates at 29MHz and is designed to image prostate tissue at a microscopic level. The source describes it as providing a 300 per cent resolution improvement over conventional ultrasound. Greater image detail can help urologists see and target regions of interest when planning a biopsy.
Imaging alone, however, does not remove the need for clinical interpretation. The next step is to use software that can analyse the images and help flag areas where cancer may be present. This is where machine learning is being investigated as a decision-support tool.
How AI supports prostate cancer diagnosis
Exact Imaging partnered with Cambridge Consultants to develop artificial intelligence for analysing ExactVu images. The AI was trained exclusively on images produced by the ExactVu system. Its intended role is to assist with identifying areas that may be suspicious, helping clinicians decide which regions to consider for biopsy.
The system uses an ensemble of deep neural networks. Rather than relying on a single model, an ensemble brings together multiple networks, with the aim of drawing on their analyses to support the output. According to the project team, this architecture was selected to provide high specificity while working with a relatively small dataset.
Specificity refers to how effectively a diagnostic approach identifies cases that do not have the target condition. In practice, an AI result should be understood as information to support a clinician’s judgement, not as a diagnosis on its own. The reported work describes a proof of concept and early testing, rather than a finished clinical product.
Working with limited medical imaging data
Deep-learning systems are often associated with large training datasets, but healthcare applications may have fewer labelled examples available. In this project, the ultrasound technology was relatively new, and the dataset consisted of images from a few hundred patients.
That constraint shaped the development approach. The team used its experience in medical diagnostics to design an ensemble of neural networks suited to the available data. The work reflects a broader technical challenge in medical AI: developing methods that can learn useful patterns without relying on the enormous datasets often used in other applications.
Training the AI exclusively on ExactVu images also defines the context in which its analysis was developed. The described system is specific to that imaging source; the source does not establish that it can be applied interchangeably to images from other ultrasound equipment.
Potential value for biopsy planning
If further development and validation support its use, an AI analysis tool could help urologists focus attention on regions that merit consideration during biopsy planning. The project team said the approach could enable more informed decisions about which areas to sample. That is a potential application, not a claim that the system has already improved patient outcomes or replaced existing clinical procedures.
The whitepaper also notes that the machine-learning approach is faster and less computationally intensive than traditional statistical techniques. Early testing of the proof of concept was described as promising. Further work would be needed to establish how the system performs in routine clinical settings and how clinicians should incorporate its outputs into their decisions.
Conclusion: combining imaging detail with clinical judgement
High-resolution prostate ultrasound and AI address two related parts of the diagnostic process: obtaining more detailed images and analysing them for regions that may be suspicious. The proposed combination could support biopsy targeting, particularly where conventional ultrasound has difficulty differentiating tissue.
For technical and clinical decision-makers, the key implication is that AI can be developed as an aid to image interpretation even when medical datasets are relatively small. The system described remains a proof of concept, and its practical value depends on further validation. Its purpose is to inform urologists’ decisions—not to replace clinical expertise or independently determine a diagnosis.