Prostate Lesion Detection and Salient Feature Assessment Using Zone-Based Classifiers

Abstract: Multi-parametric magnetic resonance imaging (mpMRI) has a growing role in detecting prostate cancer lesions. Thus, it is pertinent that medical professionals who interpret these scans reduce the risk of human error by using computer-aided detection systems. The variety of algorithms used in system implementation, however, has yielded mixed results. Here we investigate the best machine learning classifier for each prostate zone. We also discover salient features to clarify the models' classification rationale. Of the data provided, we gathered and augmented T2 weighted images and apparent diffusion coefficient map images to extract first through third order statistical features as input to machine learning classifiers. For our deep learning classifier, we used a convolutional neural net (CNN) architecture for automatic feature extraction and classification. The interpretability of the CNN results was improved by saliency mapping to understand the classification mechanisms within. Ultimately, we concluded that effective detection of peripheral and anterior fibromuscular stroma (AS) lesions depended more on statistical distribution features, whereas those in the transition zone (TZ) depended more on textural features. Ensemble algorithms worked best for PZ and TZ zones, while CNNs were best in the AS zone. These classifiers can be used to validate a radiologist's predictions and reduce inter-reader variability in patients suspected to have prostate cancer. The salient features reported in this study can also be investigated further to better understand hidden features and biomarkers of prostate lesions with mpMRIs.

• The paper demonstrates that employing zone-based classifiers enhances prostate lesion detection by tailoring machine learning pipelines to distinct prostate regions.
• Results reveal that ensemble methods outperform in the peripheral and transition zones, while CNN excels in the anterior fibromuscular stroma with high sensitivity.
• The study utilizes advanced mpMRI feature extraction techniques to improve interpretability and accuracy in prostate cancer diagnostics.

Prostate Lesion Detection and Salient Feature Assessment Using Zone-Based Classifiers

Introduction

The paper "Prostate Lesion Detection and Salient Feature Assessment Using Zone-Based Classifiers" (2208.11522) addresses the application of machine learning and deep learning models for detecting prostate lesions using multi-parametric magnetic resonance imaging (mpMRI). This is critical due to the limitations of traditional prostate cancer diagnostic methods, such as PSA tests and digital rectal exams, which often lead to false positives and negatives. The paper investigates the efficacy of different classifiers tailored specifically for various prostate zones and explores salient features to elucidate the models' decision-making processes.

Methodology

The study utilizes mpMRI data, focusing on T2-weighted images and apparent diffusion coefficient maps for feature extraction. The prostate is divided into three distinct zones: peripheral zone (PZ), transition zone (TZ), and anterior fibromuscular stroma (AS). Each zone's samples are independently processed through classification pipelines involving both traditional machine learning methods and a custom convolutional neural network (CNN) architecture.

Data Processing and Feature Extraction: Images undergo intensity range standardization and data augmentation to ensure uniformity across training and test datasets. Feature extraction techniques incorporate first through third order statistical features—including Haralick and Tamura texture features—from T2WI and ADC images.

Classifiers Used: The study evaluates multiple classifiers: Logistic Regression, Support Vector Machine (SVM), Random Forest, XGBoost, and CNN. Each classifier is optimized via hyperparameter tuning specific to the prostate zones.

Experimental Results

The results indicate that classifier performance varies significantly across different prostate zones, with ensemble methods like Random Forest and XGBoost performing well in PZ and TZ, while CNNs excel in the AS zone.

Peripheral Zone: Salient features include first and second order statistics, with the ensemble methods demonstrating superior predictive power owing to well-defined mass characteristics on MRI.

Transition Zone: The complexity of lesion texture drives the need for advanced texture feature extraction, where XGBoost performs best by leveraging strong correlated texture gradient information.

Anterior Fibromuscular Stroma Zone: The CNN architecture is markedly effective, capitalizing on high sensitivity and specificity. The saliency map analysis suggests CNN robustness in identifying distinct lesion shapes and low-signal intensity regions.

Implications and Future Work

The insights gleaned from this study have substantial implications for clinical prostate cancer diagnostics. Implementing these advanced computer-aided detection systems can reduce inter-observer variability and potentially enhance early diagnosis accuracy. Future research could expand datasets to further refine model training processes and validate these findings across more diversified clinical settings. Enhanced pre-training of CNNs on larger datasets could bolster lesion recognition capabilities, while zone-specific salient features might aid in the development of precise, zone-targeted diagnostic AI tools.

Conclusion

The paper successfully explores classifiers tailored to the specific anatomical zones of the prostate and highlights the effective use of deep learning in medical imaging. The elucidation of salient features offers deeper insights into the classification processes, providing a foundation for more robust and interpretable diagnostic systems in the future. Further research could look into integrating these models into clinical workflows to enhance diagnostic accuracy and patient prognosis.