Evaluation of a radiomics nomogram derived from Fluoride-18 PSMA-1007 PET/CT for risk stratification in newly diagnosed prostate cancer

Zhuonan Wang, Yunxuan Li, Anqi Zheng, Jungang Gao, Yuan Wang, Cong Shen, Lu Bai, Xiaoyi DuanFrontiers in Oncology 2022 · DOI

TL;DR

This study explores using radiomics features from 18F-PSMA-1007 PET/CT scans to predict prostate cancer (PCa) aggressiveness (Gleason Score) in newly diagnosed patients. They found that a radiomics signature, combined with PSA levels and metastasis status, can effectively stratify PCa risk, potentially reducing the need for unnecessary biopsies.

What Problem Does This Paper Address?

Prostate cancer is a leading cause of cancer death, but its behavior is highly variable. Current methods for assessing risk, like PSA levels and Gleason Scores (GS), have limitations. PSA can be elevated for non-cancerous reasons, leading to unnecessary biopsies. While GS is a key indicator of aggressiveness, obtaining it often requires a biopsy, which carries risks and costs. There's a need for non-invasive, objective methods to accurately assess PCa malignancy and guide treatment decisions. 18F-PSMA-1007 PET/CT is a promising imaging technique for PCa, and this paper investigates if radiomics – extracting quantitative features from these images – can provide this much-needed objective risk stratification.

Methods

The researchers retrospectively analyzed data from 161 PCa patients. They extracted a large number of radiomic features (944) from 18F-PSMA-1007 PET/CT images using the pyRadiomics tool. To identify the most predictive features for Gleason Score (GS), they employed the Least Absolute Shrinkage and Selection Operator (LASSO) logistic regression, which effectively reduces dimensionality and selects relevant features. This resulted in a "radiomics signature" of 30 features.

They then developed a radiomics nomogram, a graphical tool for predicting outcomes, by combining this radiomics signature with clinical factors: Prostate-Specific Antigen (PSA) levels and metastasis status. This nomogram was built using multivariable logistic regression. The performance of both the radiomics signature alone and the combined nomogram was evaluated using Receiver Operating Characteristic (ROC) curves and Area Under the Curve (AUC) metrics.

Datasets

The study utilized a dataset of 161 prostate cancer patients diagnosed between September 2020 and June 2021 at a single center. Patients were selected based on biopsy or radical prostatectomy proven PCa and had undergone 18F-PSMA-1007 PET/CT scans. The dataset was split into training (112 patients) and testing (49 patients) cohorts for model development and validation. No new benchmark datasets were introduced.

Key Insights and Ideas

  • Radiomics as a Non-Invasive Biomarker: The study highlights the potential of radiomics from 18F-PSMA-1007 PET/CT to act as a non-invasive "imaging biomarker" for PCa aggressiveness, complementing traditional markers like PSA and GS.
  • Predicting Gleason Score: The radiomics signature derived from PET/CT images showed a significant association with both PSA levels and Gleason Score, suggesting its ability to capture underlying tumor biology.
  • Nomogram for Integrated Risk Assessment: Combining radiomics features with clinical data (PSA and metastasis status) into a nomogram provides a more comprehensive risk assessment tool.
  • Reducing Unnecessary Biopsies: By offering a more objective measure of malignancy, this approach could help clinicians better identify patients who truly need further invasive procedures, thereby reducing unnecessary biopsies.

Key Results

  • A radiomics signature consisting of 30 features was developed and found to be significantly associated with PSA level and Gleason Score (P < 0.001) in both training and validation cohorts.
  • The radiomics nomogram, incorporating the radiomics signature, PSA level, and metastasis status, achieved an AUC of 0.719 for predicting Gleason Score in the test cohort.
  • The combined clinical-radiomic nomogram demonstrated a similar benefit in discriminating Gleason Score as using radiomics features alone.
  • Semi-quantitative PET/CT parameters (SUVmin, SUVmax, SUVmean) were significantly higher in patients with higher PSA levels (≥20 ng/mL) and higher Gleason Scores (≥8).

Limitations and Future Work

The authors acknowledge several limitations:

  • Single-Center Study: The findings are based on data from a single institution, which may limit generalizability.
  • Retrospective Design: The study design is retrospective, which can introduce biases.
  • Manual Segmentation: The manual segmentation of tumors for radiomics extraction is subjective and can lead to measurement errors.

Future work suggested includes:

  • External Validation: The results need to be validated on larger, multi-center datasets to confirm their robustness and clinical applicability.
  • Automatic Segmentation: Developing and implementing automatic segmentation methods could improve accuracy and reduce inter-observer variability.
  • Integration with CT Radiomics: Incorporating CT radiomics features alongside PET radiomics could potentially further enhance diagnostic performance.
  • Combining Imaging and Pathology: Future research should aim to integrate PET/CT radiomics findings with pathological data for a more comprehensive understanding of tumor biology and improved diagnostic accuracy.

Summary updated

Abstract

The <sup>18</sup>F-PSMA-1007-PET/CT radiomics signature can be used to facilitate preoperative individualized prediction of GS; incorporating the radiomics signature, PSA level, and metastasis status had similar benefits to those of utilizing the PET/CT radiomics features alone.
BibTeX
@inproceedings{wang2022evaluation,
  author = {Zhuonan Wang and Yunxuan Li and Anqi Zheng and Jungang Gao and Yuan Wang and Cong Shen and Lu Bai and Xiaoyi Duan},
  title = {{Evaluation of a radiomics nomogram derived from Fluoride-18 PSMA-1007 PET/CT for risk stratification in newly diagnosed prostate cancer}},
  booktitle = {Frontiers in Oncology},
  year = {2022},
  doi = {10.3389/fonc.2022.1018833},
}
Figure