Deep Learning-Based Classification and Targeted Gene Alteration Prediction from Pleural Effusion Cell Block...

Area of Science:

• Oncology
• Pathology
• Artificial Intelligence

Background:

• Pleural effusion is crucial for diagnosing advanced cancers, but cytopathologist shortages and high gene detection costs hinder its use.
• Deep learning offers a potential solution to these diagnostic challenges in cytopathology.

Purpose of the Study:

• To develop and evaluate a deep learning model for diagnosing pleural effusion and predicting genetic alterations.
• To assess the model's ability to differentiate benign from malignant effusions and identify primary cancer locations.

Main Methods:

• A retrospective analysis of 1321 consecutive pleural effusion cases was performed.
• A deep learning model was trained to diagnose effusion type, identify metastatic cancer origins, and predict gene alterations for targeted therapy.

Main Results:

• The model achieved high accuracy in distinguishing benign and malignant pleural effusions (0.932 AUC).
• It also demonstrated strong performance in identifying the primary location of common metastatic cancers (0.910 AUC).
• Reasonable AUCs were obtained for predicting genetic alterations, including ALK fusion (0.869), KRAS mutation (0.804), EGFR mutation (0.644), and NONE alteration (0.774).

Conclusions:

• Deep learning is feasible and beneficial for assisting cytopathological diagnosis in clinical settings.
• AI-powered tools can help overcome limitations in cytopathology resources and improve cancer diagnosis and treatment.
• The model shows potential for aiding in the identification of actionable genetic alterations for targeted cancer therapy.