Clinical Scorecard: Utilizing Deep Learning for Predicting Tumor Origin and Malignancy in Cytology and Histopathology Whole Slide Images

At a Glance

Key Highlights

• MAMILE-UNI achieved high accuracy, AUROC, sensitivity, specificity, and F1 scores in detecting malignancy from pleural and ascitic cytology WSIs.
• The model accurately identified the origin of cancer from both cytology smears and histopathological slide images.
• Statistical validation with Fisher’s exact test confirmed the significance of model predictions (p < 0.001).

Guideline-Based Recommendations

Diagnosis

• Use pleural and ascitic cytology as essential tools for diagnosing metastatic cancer and predicting tumor origin.
• Incorporate deep learning models like MAMILE-UNI to improve diagnostic accuracy and reduce observer variability.

Management

• Follow established guidelines for managing malignant pleural effusions and ascites (e.g., British Thoracic Society 2010).
• Utilize cell block preparations alongside cytology smears to enhance diagnostic yield.

Monitoring & Follow-up

• Regularly assess diagnostic performance metrics such as sensitivity, specificity, and accuracy when implementing AI tools.
• Validate AI model predictions statistically to ensure clinical reliability.

Risks

• Be aware of limitations in microscopic observation alone due to low accuracy and observer variability.
• Consider the need for data-efficient and validated AI frameworks to minimize diagnostic errors.

Patient & Prescribing Data

Patients undergoing cytological evaluation for pleural and ascitic effusions with suspected malignancy

Deep learning-assisted diagnosis can guide targeted management by accurately identifying malignancy and tumor origin, potentially improving treatment decisions.

Clinical Best Practices

• Combine cytology smear and cell block whole slide imaging for comprehensive evaluation.
• Implement validated deep learning frameworks to support pathologists in malignancy detection and tumor origin prediction.
• Ensure open access to datasets and code to facilitate reproducibility and clinical adoption.
• Adhere to established cytology and pathology guidelines for sample collection, preparation, and interpretation.

References

• British Thoracic Society Pleural Disease Guideline 2010
• The Cancer Genome Atlas (TCGA) via Genomic Data Commons
• MAMILE-UNI GitHub Repository
• Kassirian et al. Diagnostic sensitivity of pleural fluid cytology in malignant pleural effusions: systematic review and meta-analysis
• Srinivasan et al. Indian Academy of Cytologists guidelines for serous effusion fluid samples