• HOME
  • NEWS
  • EXPLORE
    • CAREER
      • Companies
      • Jobs
    • EVENTS
    • iGEM
      • News
      • Team
    • PHOTOS
    • VIDEO
    • WIKI
  • BLOG
  • COMMUNITY
    • FACEBOOK
    • INSTAGRAM
    • TWITTER
Tuesday, November 17, 2026
BIOENGINEER.ORG
No Result
View All Result
  • Login
  • HOME
  • NEWS
  • EXPLORE
    • CAREER
      • Companies
      • Jobs
        • Lecturer
        • PhD Studentship
        • Postdoc
        • Research Assistant
    • EVENTS
    • iGEM
      • News
      • Team
    • PHOTOS
    • VIDEO
    • WIKI
  • BLOG
  • COMMUNITY
    • FACEBOOK
    • INSTAGRAM
    • TWITTER
  • HOME
  • NEWS
  • EXPLORE
    • CAREER
      • Companies
      • Jobs
        • Lecturer
        • PhD Studentship
        • Postdoc
        • Research Assistant
    • EVENTS
    • iGEM
      • News
      • Team
    • PHOTOS
    • VIDEO
    • WIKI
  • BLOG
  • COMMUNITY
    • FACEBOOK
    • INSTAGRAM
    • TWITTER
No Result
View All Result
Bioengineer.org
No Result
View All Result
Home NEWS Science News Health

PSMA PET Radiomics Machine Learning Models Predict Prostate Cancer Genomic Alterations

by
November 13, 2026
in Health
Reading Time: 5 mins read
0
PSMA PET Radiomics Machine Learning Models Predict Prostate Cancer Genomic Alterations

PSMA PET Radiomics Machine Learning Models Predict Prostate Cancer Genomic Alterations

Share on FacebookShare on TwitterShare on LinkedinShare on RedditShare on Telegram

A retrospective pilot study evaluating patients with metastatic prostate cancer demonstrated that machine learning algorithms combining quantitative PET imaging features and clinical metrics can non-invasively predict specific somatic genomic alterations. By examining primary prostatic tissue and metastatic bone lesions, the study evaluated computational models designed to anticipate alterations in key cancer-related genes, including TP53, TMPRSS2, and PTEN, pointing toward potential non-invasive imaging surrogates to inform precision oncology workflows.

Metastatic prostate cancer presents considerable biological heterogeneity and a complex mutational landscape that strongly influences diagnostic classification, prognosis, and systemic treatment pathways. Precision oncology strategies increasingly rely on tissue-based genomic profiling to guide therapy, yet comprehensive genetic testing involves notable limitations, including high costs, invasive procedures, and extended laboratory turnaround times. In advanced metastatic stages, genetic alterations frequently associated with aggressiveness and therapeutic resistance, such as TP53, PTEN, and TMPRSS2, play crucial prognostic roles. In parallel, prostate-specific membrane antigen positron emission tomography has become a central imaging modality in prostate cancer care because of its documented high sensitivity and specificity. The research team aimed to establish whether computational radiomic analysis could link non-invasive imaging phenotypes to underlying tumor genomic signatures, potentially offering surrogate biological insights when tissue sampling is unfeasible or when rapid clinical stratification is necessary.

The study cohort comprised 14 patients diagnosed with metastatic prostate cancer who underwent [18 F]PSMA-1007 PET/CT scans following standardized institutional imaging protocols at the National Cancer Institute in Mexico City. Across the 14 patients, nuclear medicine physicians manually segmented a total of 46 distinct lesions, encompassing the prostate gland and metastatic bone lesions while remaining blinded to corresponding genomic data. Clinical records indicated that 8 patients, or 57 percent, presented with low-volume metastatic disease, while 6 patients, representing 42 percent, had high-volume metastatic disease defined by visceral involvement or four or more bone lesions with at least one lesion located beyond the vertebral bodies and pelvis. The median patient age at diagnosis was 57.7 years, spanning a range from 47 to 62 years, and all participants were of Mexican origin. Median prostate-specific antigen at diagnosis was 991.8 ng/mL with a broad range between 9.2 and 7414 ng/mL, and the median Gleason score was 9, ranging between 7 and 9. All 14 patients had received androgen deprivation therapy, including LHRH analogues, enzalutamide, or abiraterone, while 6 patients received chemotherapy and 7 received radiotherapy.

Somatic genomic alterations were characterized from formalin-fixed, paraffin-embedded prostate tissue samples using the comprehensive FoundationOne CDx genomic profiling assay. This commercial sequencing panel analyzed a broad spectrum of cancer-related genes along with genomic signatures such as tumor mutational burden, microsatellite instability, and loss of heterozygosity. In this cohort, testing detected somatic alterations across several genes, including PIK3CA, PTEN, SOX2, PRKCI, TERC, TMPRSS2, CDKN2A, ATM, BRCA1, BRCA2, RNF43, MAP2K1, CTNNB1, AKT1, STAG2, TP53, SPEN, SPOP, FLCN, and SRC. All 14 patients exhibited microsatellite stability and low tumor mutational burden scores ranging from 0 to 4 mutations per megabase. Due to sample size limitations and the low individual frequency of many mutations, alterations in genes other than TP53, TMPRSS2, and PTEN were consolidated into a single composite category designated as OTHER for subsequent predictive modeling.

For image processing, the researchers utilized LifeX, an image analysis platform compliant with the Image Biomarker Standardization Initiative guidelines. Before feature extraction, standard preprocessing steps were executed, including resampling image voxels to isotropic dimensions, discretizing image intensity into 64 bins, and applying grey-level normalization. From each segmented volume of interest, nuclear medicine specialists extracted 86 quantitative radiomics features, covering first-order statistics, intensity histogram metrics, and three-dimensional shape descriptors. These imaging metrics were integrated with four baseline clinical variables: Gleason score, prostate-specific antigen concentration, International Society of Urological Pathology grade, and tumor mutational burden, establishing the primary feature pool evaluated in subsequent computational modeling pipelines.

The machine learning workflow was developed using the matRadiomics software environment, testing six distinct algorithmic approaches: Discriminant Analysis, Support Vector Machines, K-Nearest Neighbours, Neural Networks, Random Forest, and Boosting utilizing the AdaBoostM1 algorithm. To prevent data leakage and evaluate generalizability, the investigators implemented a five-times repeated holdout framework with an 80 percent training and 20 percent testing data split. Dimensionality reduction within each training partition was performed using the Least Absolute Shrinkage and Selection Operator, known as LASSO, followed by five-fold cross-validation exclusively on the training sets for hyperparameter optimization and model fitting. The held-out test partitions were reserved strictly for final model evaluation, measuring accuracy, area under the receiver operating characteristic curve, sensitivity, specificity, precision, and F1-score, alongside feature stability tracking across repetitions.

When assessing the combined clinical-radiomics models across specific mutations, classification performance varied according to the genomic target. For TP53 alterations, the Boosting model delivered the top overall performance, registering an accuracy of 91.11 percent, an area under the curve of 93.10 percent, a true positive rate of 95.00 percent, a true negative rate of 90.07 percent, a positive predictive value of 87.00 percent, and an F1-score of 89.29 percent. Highly recurrent features selected in 100 percent of repetitions for this combined TP53 model included the intensity histogram metrics EntropyLog10 and MinimumGreyLevel, alongside serum prostate-specific antigen levels and tumor mutational burden. For predicting TMPRSS2 mutations, the Support Vector Machine model demonstrated the strongest overall combined metrics, attaining an accuracy of 84.44 percent, an area under the curve of 87.00 percent, an F1-score of 83.33 percent, and a specificity of 74.00 percent, with MaximumGreyLevel being selected in 80 percent of repetitions.

In evaluating PTEN mutations within the combined configuration, the Boosting algorithm achieved the best diagnostic results, posting an accuracy of 80.00 percent, an area under the curve of 83.50 percent, a sensitivity of 75.00 percent, and a specificity of 84.00 percent. The most frequently selected variables for this model, each appearing in 100 percent of cross-validation iterations, were Gleason score, Minimum Intensity, Discretized Tumor Volume, and tumor mutational burden. For the composite OTHER mutation category, the Support Vector Machine algorithm led the combined analysis, reaching an accuracy of 77.78 percent, an area under the curve of 74.44 percent, and an F1-score of 84.80 percent, with Maximum Histogram Gradient Grey Level and Voxel Count emerging as the most frequently retained features across 60 percent of runs.

A comparative analysis contrasting modality configurations indicated that no single setup—radiomics-only, clinical-only, or multimodal combined—uniformly outperformed the others across all targets. In radiomics-only models, Discriminant Analysis achieved an F1-score of 92.00 percent for TP53, while Boosting reached 71.19 percent for PTEN and Discriminant Analysis reached 81.10 percent for OTHER mutations. In clinical-only architectures, a Neural Network yielded an F1-score of 100.00 percent for TP53, though the authors explicitly cautioned that this flawless training performance likely reflects data overfitting that requires external validation. For PTEN, however, combining modalities showed a substantial synergistic gain: the combined Neural Network achieved an F1-score of 80.99 percent, vastly exceeding both the radiomics-only Boosting model at 71.19 percent and the clinical-only K-Nearest Neighbours model at 55.32 percent, underscoring that multivariable data integration is vital for complex genomic alterations.

The authors emphasized several fundamental limitations inherent to their exploratory pilot investigation, particularly the constrained cohort size of 14 patients and 46 total lesions. Because multiple lesions from individual patients were analyzed as independent samples without hierarchical or mixed-effects adjustments, the effective sample size may have been artificially inflated, potentially introducing bias into the performance estimates. Furthermore, models did not adjust for lesion volume as a covariate, leaving open the possibility that certain shape and textural radiomic descriptors capture volumetric effects rather than intrinsic biology. The researchers concluded that while their findings demonstrate the initial feasibility of PSMA PET-derived radiogenomics in metastatic prostate cancer, future validation in larger multicenter cohorts with automated segmentation and standardized genomic sequencing is essential before clinical translation can occur.

Subject of Research: Feasibility of machine learning models combining PSMA PET radiomics and clinical data to predict genomic alterations in metastatic prostate cancer

Article Title: Radiomics-based machine learning models for predicting genomic alterations in metastatic prostate cancer using PSMA PET imaging: a pilot study

Article References: Scavuzzo, A., Pasini, G., Perez, O. G., Jimenez Rios, M. A., Hernández, M. A., Pedrero-Piedras, R., Pérez Montiel, M. D., Moreno, N. S., Russo, G., & Stefano, A. (2025). Radiomics-based machine learning models for predicting genomic alterations in metastatic prostate cancer using PSMA PET imaging: a pilot study. EJNMMI Reports, 9(1), Article 42. https://doi.org/10.1186/s41824-025-00280-6

Image Credits: AI Generated

DOI: 10.1186/s41824-025-00280-6

Keywords: metastatic prostate cancer, PSMA PET CT, radiomics, machine learning, genomic alterations, TP53, PTEN, TMPRSS2, precision oncology, radiogenomics, FoundationOne CDx, LifeX, biomarkers, feature selection, predictive modeling, pilot study, tumor heterogeneity

News Source: Nathaniel Bowman. (November 13, 2026). PSMA PET Radiomics Machine Learning Models Predict Prostate Cancer Genomic Alterations. Scienmag.

Tags: biomarkersFeature SelectionFoundationOne CDxgenomic alterationsLifeXMachine Learningmetastatic prostate cancerPilot Studyprecision oncologypredictive modelingPSMA PET/CTPTENRadiogenomicsradiomicsTMPRSS2TP53tumor heterogeneity
Share12Tweet7Share2ShareShareShare1

Related Posts

Adapted Virtual Presence Tool Proves Feasible in Pilot Manikin Simulation Courses

Adapted Virtual Presence Tool Proves Feasible in Pilot Manikin Simulation Courses

November 17, 2026
German Network Launches National Multi-Center Protocol for Longitudinal Stroke Research

German Network Launches National Multi-Center Protocol for Longitudinal Stroke Research

November 17, 2026

Neurological Paraneoplastic Syndromes Often Precede Gynecological and Breast Cancer Diagnoses

November 17, 2026

97-Plex Y-STR System Reveals Distinct Paternal Lineages in Two Chinese Groups

November 17, 2026

POPULAR NEWS

  • Structured IDEA Expert Elicitation Framework Enhances Design of Rare Disease Trials

    Structured IDEA Expert Elicitation Framework Enhances Design of Rare Disease Trials

    29 shares
    Share 12 Tweet 7
  • Syndesmotic Fixation Speeds Early Ankle Recovery but Matches Non-Fixation at Two Years

    29 shares
    Share 12 Tweet 7
  • German Nationwide Survey Identifies Workplace Bullying Patterns Across Industries and Age Groups

    29 shares
    Share 12 Tweet 7
  • TCAD Simulations Identify Efficiency Gains for Sustainable CZTS Solar Cells

    29 shares
    Share 12 Tweet 7

About

We bring you the latest biotechnology news from best research centers and universities around the world. Check our website.

Follow us

Recent News

Adapted Virtual Presence Tool Proves Feasible in Pilot Manikin Simulation Courses

German Network Launches National Multi-Center Protocol for Longitudinal Stroke Research

Neurological Paraneoplastic Syndromes Often Precede Gynecological and Breast Cancer Diagnoses

Subscribe to Blog via Email

Enter your email address to subscribe to this blog and receive notifications of new posts by email.

Join 88 other subscribers
  • Contact Us

Bioengineer.org © Copyright 2023 All Rights Reserved.

Welcome Back!

Login to your account below

Forgotten Password?

Retrieve your password

Please enter your username or email address to reset your password.

Log In
No Result
View All Result
  • Homepages
    • Home Page 1
    • Home Page 2
  • News
  • National
  • Business
  • Health
  • Lifestyle
  • Science

Bioengineer.org © Copyright 2023 All Rights Reserved.