Malignant pleural mesothelioma classification and survival prediction with CT imaging using ResNet

Area of Science:

• Radiology and Medical Imaging
• Artificial Intelligence in Medicine
• Oncology

Background:

• Malignant pleural mesothelioma (MPM) diagnosis is challenging due to its rarity and overlap with metastatic pleural disease (MPD).
• Accurate differentiation and prognostic prediction are crucial for effective patient management.
• Current diagnostic methods can be invasive and time-consuming.

Purpose of the Study:

• To develop and validate a deep learning model for differentiating MPM from MPD using CT scans.
• To assess the prognostic capability of the deep learning model for predicting overall survival in MPM patients.
• To compare the deep learning model's performance against a clinical benchmark.

Main Methods:

• A retrospective study utilized CT scans from 385 patients (85 MPM, 290 MPD).
• A ResNet-3D-18 model was trained to distinguish MPM from MPD, with validation on an independent cohort.
• Deep learning features were analyzed for prognostic utility using a random forest classifier.

Main Results:

• The ResNet-3D-18 model achieved high accuracy in differentiating MPM from MPD, with AUCs of 0.972 (training) and 0.840 (test).
• The deep learning approach demonstrated superior sensitivity (0.867) compared to the clinical model (0.533) in the test cohort.
• The random forest classifier predicted MPM overall survival with an AUC of 0.829.

Conclusions:

• The ResNet-3D-18 model effectively differentiates MPM from MPD.
• Morphological features identified by deep learning contain prognostic information for MPM.
• This AI-driven approach offers a non-invasive tool for early diagnosis and personalized prognosis in MPM.