Multi-task deep learning based CT imaging analysis for COVID-19 pneumonia: Classification and segmentation.

Area of Science:

• Medical Imaging
• Artificial Intelligence
• Computer-Aided Diagnosis

Background:

• Chest CT imaging is crucial for diagnosing COVID-19 pneumonia.
• Accurate segmentation of lesions and classification of COVID-19 are essential for patient management.
• Limited data availability poses challenges for developing robust deep learning models.

Purpose of the Study:

• To develop an automatic classification and segmentation tool for COVID-19 pneumonia using chest CT images.
• To propose a multitask deep learning model for joint identification and segmentation of COVID-19 lesions.
• To leverage related tasks and small datasets for improved segmentation and classification performance.

Main Methods:

• A novel multitask deep learning architecture with a common encoder and task-specific decoders was designed.
• The model jointly performs segmentation, classification, and reconstruction tasks.
• Evaluation involved a dataset of 1369 patients, including COVID-19, normal, lung cancer, and other pathologies.

Main Results:

• The proposed model achieved a Dice coefficient exceeding 0.88 for lesion segmentation.
• The classification performance demonstrated an area under the ROC curve greater than 97% for COVID-19 detection.
• The multitask approach effectively addressed challenges associated with small datasets.

Conclusions:

• The developed deep learning tool shows significant promise for automated screening and assessment of COVID-19 pneumonia.
• The multitask learning strategy enhances performance and data efficiency in medical image analysis.
• This approach can aid clinicians in evaluating pneumonia severity and monitoring patient recovery.