Deep Learning-Based CT Imaging in Diagnosing Myeloma and Its Prognosis Evaluation

Area of Science:

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

Background:

• Multiple myeloma diagnosis relies heavily on imaging.
• Accurate segmentation and classification of myeloma lesions on CT scans are crucial for early detection and treatment planning.
• Deep learning offers potential for improving the accuracy of medical image analysis.

Purpose of the Study:

• To evaluate the effectiveness of deep learning models (U-Net and Faster RCNN) for segmenting and labeling myeloma lesions in CT images.
• To compare the prognostic outcomes of patients with myeloma treated with bortezomib-based versus non-bortezomib chemotherapy regimens.

Main Methods:

• An improved U-Net model was utilized for segmenting CT images of 186 patients with suspected myeloma.
• The Faster RCNN model was employed to label myeloma lesions, achieving a 99% classification accuracy.
• Patients were divided into bortezomib (n=128) and non-bortezomib (n=58) groups for comparative analysis of treatment effects.

Main Results:

• The U-Net model achieved effective segmentation of myeloma on CT images.
• The Faster RCNN model accurately labeled lesion areas with high classification accuracy.
• The non-bortezomib group exhibited significant improvements, including muscle growth and reductions in key myeloma biomarkers (e.g., M protein, β2MG, WBC), compared to the bortezomib group (P < 0.05).

Conclusions:

• Deep learning models enhance the accuracy of myeloma detection through improved CT image segmentation and classification.
• Both chemotherapy regimens benefit patient prognosis, but non-bortezomib regimens demonstrate superior efficacy.