AI model identifies tumor volume as risk factor for prostate cancer metastasis, recurrence
Click Here to Manage Email Alerts
Key takeaways:
- An AI model showed tumor size to be an independent risk factor for prostate cancer metastasis and recurrence.
- AI can significantly reduce the time clinicians spend outlining prostate tumors on MRI.
An AI model revealed tumor size to be an independent factor when determining risk for prostate cancer recurrence or metastasis, according to results of a retrospective study.
AI also “reliably and consistently” detected and localized the area of concern within the prostate, David D. Yang, MD, radiation oncologist at Brigham and Women’s Hospital and Dana-Farber Cancer Institute, as well as instructor at Harvard Medical School, told Healio.
“With this AI approach to MRI, because it independently predicts for the aggressiveness of the cancer, it may help physicians make more personalized treatment recommendations,” Yang said.
Background and methods
MRI has become standard for men with localized prostate cancer, helping clinicians distinguish tumors that may be clinically aggressive from those that could be indolent, Yang said.
However, clinicians rarely see the images the exact same way.
“If you were to give the same image to 10 different clinicians and have them outline the area of cancer view, they would do it 10 different ways. There is going to be interobserver variability,” Yang said.
The process also can be time consuming, Yang added.
It takes 10 to 15 minutes to outline each tumor on an MRI. AI can do it in seconds.
“Ultimately, we see these tools as ones to help human clinicians ... not replace them,” Yang said.
Yang and colleagues trained an AI model with hand-outlined MRIs of prostate cancer.
They evaluated the model on 732 consecutive patients diagnosed with cT1c-T3bN0M0 prostate cancer at Brigham and Women’s Hospital or Dana-Farber Cancer Institute between January 2021 and August 2023.
The cohort consisted of men who had MRI prior to radiation therapy (n = 438; median age, 68 years; interquartile range, 62-73; median follow-up, 6.9 years) or radical prostatectomy (n = 294 men; median age, 61 years; interquartile range, 56-66; median follow-up, 5.5 years).
Researchers randomly divided men who had radiation therapy into two groups — a cross-validation cohort and a test cohort.
Prognostic information derived from AI-determined tumor volume served as the primary endpoint.
Results and next steps
Compared with clinician-outlined MRIs, the AI model correctly marked 87% of tumor edges in the cross-validation radiation therapy group, 84% in the test radiation therapy group and 85% in the radical prostatectomy group.
“The areas the AI model was not picking up were ones with lesions that were less obvious on the scans,” Yang said. “What we think may be happening is that the model may not be picking up the areas that are biologically less aggressive, but they are picking up more of the areas that are biologically more aggressive.”
The AI model also identified tumor size as an independent risk factor for recurrence or metastasis in the overall radiation therapy cohort (adjusted HR = 1.09 per milliliter increase; 95% CI, 1.04-1.15) and the radical prostatectomy group (adjusted HR = 1.22 per milliliter increase; 95% CI, 1.08-1.39).
These data, combined with current risk stratification from the National Comprehensive Cancer Network, could allow clinicians to make more informed treatment decisions, Yang added.
Researchers acknowledged study limitations, including its single-institution nature, which could prevent the algorithm from identifying tumors on images produced with different setting configurations.
More research is needed to determine if AI can identify other image characteristics that could impact risk predictions, such as variations in color intensities and the appearance both around and within the tumor, Yang said.
“We really are at the beginning of how AI may be helpful for patients with prostate cancer,” he added. “By seeing patterns that people cannot easily see, it may allow us to be more precise and make more personalized treatment recommendations for our patients.”
References:
- Using AI to measure prostate cancer lesions could aid diagnosis and treatment (press release). Available at: https://www.massgeneralbrigham.org/en/about/newsroom/press-releases/using-ai-to-measure-prostate-cancer-lesions-could-aid-diagnosis-and-treatment. Published Oct. 29, 2024. Accessed Nov. 14, 2024.
- Yang DD, et al. Radiology. 2024;doi:10.1148/radiol.240041.
For more information:
David D. Yang, MD, can be reached at david_yang@dfci.harvard.edu.
Collapse
Click Here to Manage Email Alerts