How AI aids prostate cancer detection

Detecting prostate cancer with AI-enabled MRI

In May personal health platform Function Health announced it had acquired Ezra to combine Function's lab testing foundation with Ezra's FDA-cleared AI-powered MRI screening. At that time, Function Health launched a version of full-body AI that now includes the FDA-cleared AI, which the company says reduces scan time from 60 minutes to 22 minutes and lowers the price of the exam from $1,500 to $499 (without insurance and out of pocket). An AI-enabled full-body MRI exam could detect cancer before symptoms, according to Function Health.

Another type of MRI screening called high-quality prostate multiparametric magnetic resonance imaging is the "gold standard" for noninvasive imaging, according to Cacciamani. However, he added that understanding those studies requires significant time and expertise.

"AI is already transforming prostate cancer detection by making mpMRI reads faster, probably more accurate and less variable," Cacciamani said. "When compared to general full‐body MRI screening protocols, AI‐augmented, prostate‐focused imaging provides a level of detail and confidence that full‐body scans simply cannot match."

After the mpMRI images are created, AI could help radiologists make sense of the images and speed up detection of cancerous tissue, according to Cacciamani.

"In practical terms, this means that from the moment an mpMRI is acquired, an AI system can generate a detailed, three‐dimensional probability map of cancerous tissue nearly instantaneously, reducing a process that once took days (scheduling the scan, waiting for the reading, planning a biopsy) to mere hours or even minutes," Cacciamani said. "By flagging high‐confidence areas early on, AI not only speeds up the diagnostic pathway but also might minimize unnecessary biopsies when the model's certainty is extremely high."

Sometimes MRI exams could bring some limitations in screening protocols, such as finding benign cysts, small adrenal nodules and nonspecific bone lesions that cause patients to receive unnecessary biopsies and procedures, he explained. The MRI exams could also bring "variability in dedicated prostate MRI interpretation," according to Cacciamani. Varying scanner, strength, coil types and reader experience can cause these differences, he said.

However, AI brings the potential to compensate for these potential weaknesses by applying uniform algorithms, he said.

"MRI itself is neither intrinsically unreliable nor inaccurate; rather, its performance depends on the protocols used and the interpreter's experience." he said. "AI helps standardize both acquisition (through quality control metrics) and interpretation (through consistent probability mapping), making prostate MRI far more dependable."

AI cancer mapping can help predict outcomes

Avenda Health, an AI healthcare company, presented a study at the American Urological Association's 2025 annual meeting, that claimed Unfold AI, its cancer-mapping tool, is more accurate than MRI in predicting cancer spread. Unfold AI delivered a 92% accuracy rate in predicting seminal vesicle invasion, which the company says is a critical factor in prostate cancer staging and prognosis, compared with a 52% accuracy rate when a radiologist evaluates standard MRI.

While full-body MRI testing can make prostate cancer detection faster and more affordable, AI diagnosis algorithms could complement full-body MRI, according to Alan Priester, a senior data scientist at Avenda.

"Studies have proven that AI performs equivalently to, and sometimes better than, trained physicians," Priester said. "Furthermore, AI algorithms often require only a few seconds to make a prediction, hundreds of times faster than a human."

However, he expressed reservations about the full-body MRI technique.

"There are concerns that full-body MRI will tend to over-detect and overdiagnose diseases, since it can be difficult to tell the difference between real cancers and benign abnormalities," Priester said. "These false-positive diagnoses could end up costing our doctors and healthcare system a lot of time, money, and resources."

To make predictions, Avenda's technology draws on multimodal data and combines information from MRI, prostate specific antigen blood tests, and biopsies, Priester said.

"It's like using all your senses versus just one," Priester said.

With Avenda, deep-learning models use artificial "neurons" to search for varying image features. Then the model learns which features are connected to prostate cancer.

"For example, the model might learn to associate cancer with brightness, texture and anatomical location," Priester said. "The model is then able to look for those features and predict prostate cancer in new images it has never seen before. In a multimodal approach, deep learning is combined with other types of data, using other 'senses' to determine where cancer is."

When AI tools analyze a prostate mpMRI, it can generate a "voxel-level map" with info on the probability of malignancy for small elements, according to Cacciamani.

"That means a radiologist can see exactly which regions within the gland warrant further attention," Cacciamani said.

A 2024 study by Avenda and UCLA researchers published in The Journal of Urology found that by using AI to help with cancer contouring, predicting cancer size was 45 times more accurate than with conventional imaging and blood tests.

Despite the benefits, there are significant drawbacks to AI technology. In a study published in June 2024 in the American Society of Clinical Oncology Educational Book, researchers noted issues using AI for prostate cancer detection, including patient privacy, managing mistakes or bias and the ability for AI systems to adapt to clinical workflows.

"Addressing these challenges requires a holistic approach that includes enhancing data quality, improving model transparency, and establishing robust standards and ethical frameworks for AI use in oncology," researchers wrote.

ASCO researchers, however, also cited how deep learning models delivered solid performance when using imaging and pathology data to detect and grade prostate cancer. In advanced prostate cancer cases, deep learning can boost prognostication and help with clinical decision-making.

Delivering analytics

AI tools can integrate imaging features and clinical variables — including PSA kinetics, family history and previous biopsy results — to create a personalized risk score, according to Cacciamani. In addition, AI systems can use quality-control metrics to track their own performance, he said. They can spot scans with "excessive motion artifacts or poor signal‐to‐noise ratios" and monitor data for different demographics or genetic subgroups.

"Together, these analytics provide a deeply data‐driven, personalized assessment of risk and help ensure consistent accuracy across diverse patient populations," Cacciamani said.

AI models can detect the probability of cancer as well as the grade, stage and tumor size. They can provide insight on the best treatment strategies and successful treatment options, Priester said. He added that the same AI model can generate multiple analytics.

"For example, the Unfold AI model has been shown to accurately predict the boundary of a tumor, the likelihood of a tumor extending beyond the prostate and the likelihood that focal treatment will succeed," Priester said.

The future of AI in prostate cancer detection

Generative AI brings the potential for noninvasive "characterization" of tumors, according to Cacciamani. Gen AI methods of detection could consist of diffusion models or generative adversarial networks that create synthetic MRI patches with lesions that appear different. These GANs can boost the variety of training data, Cacciamani said.

"Looking ahead, these generative techniques may even infer genetic mutations or detailed histologic features from imaging alone, bringing us closer to completely noninvasive tumor characterization," Cacciamani said.

Although today's AI models typically rely on MRIs to get their data, models could potentially incorporate data like ultrasound images, pathology images and biomarkers, Priester noted.

In the next few years, look for AI-driven "virtual biopsy" to become a routine aspect of prostate cancer care, Cacciamani predicted.

"The core deep learning architectures — comprising segmentation networks, detection networks, and risk‐prediction modules — work together to produce a 'virtual biopsy' in minutes," Cacciamani said. "As AI tools evolve to incorporate blood‐based biomarkers and genetic information, they will deliver increasingly precise prognostication."