Bone tumor necrosis rate detection in few-shot X-rays based on deep learning

Area of Science:

• Medical Imaging
• Artificial Intelligence in Oncology
• Computational Pathology

Background:

• Biopsy-based necrosis rate is the standard for assessing bone tumor sensitivity and guiding chemotherapy.
• Biopsy procedures are invasive, time-consuming, and challenging with limited samples (few-shot).
• Accurate necrosis rate detection is crucial for effective bone tumor treatment planning.

Purpose of the Study:

• To develop a non-invasive method for detecting bone tumor necrosis rate using time series X-ray images.
• To address the challenge of limited data in few-shot bone tumor analysis.
• To create an efficient and accurate alternative to traditional biopsy methods.

Main Methods:

• Utilized a Generative Adversarial Network with Long Short-term Memory (GAN-LSTM) to generate synthetic time series X-ray images.
• Employed an image-to-image translation network for initial image augmentation.
• Trained a 3D-Convolutional Neural Network (3D-CNN) classification model on the augmented dataset.

Main Results:

• Successfully expanded the few-shot bone tumor X-ray dataset by 10 times.
• Achieved necrosis rate classification results comparable to the gold standard biopsy method.
• Demonstrated the efficacy of the proposed method in few-shot bone tumor necrosis rate detection.

Conclusions:

• The developed method offers an efficient, non-invasive approach for bone tumor necrosis rate assessment.
• Artificial intelligence techniques can effectively augment limited medical imaging data for improved diagnostics.
• This approach shows significant potential for clinical application in bone tumor management.