ImSpect: Image-driven self-supervised learning for surgical margin evaluation with mass spectrometry.

Area of Science:

• Oncology
• Medical Devices
• Artificial Intelligence

Background:

• Real-time surgical margin assessment is crucial for cancer patient outcomes.
• The iKnife (mass spectrometry device) shows promise for intraoperative margin detection.
• Existing iKnife research lacks integration with state-of-the-art computer vision models and datasets.

Purpose of the Study:

• To develop a novel framework (ImSpect) for enhanced surgical margin characterization using iKnife data.
• To leverage deep learning and self-supervision for improved real-time tissue analysis.
• To bridge the gap between iKnife technology and advanced computer vision techniques.

Main Methods:

• The ImSpect framework converts 1D iKnife mass spectrometry data into 2D images.
• State-of-the-art image classification networks are applied to these 2D images.
• Self-supervision is utilized to train models on large amounts of unlabeled intraoperative data.

Main Results:

• The ImSpect framework surpasses previous benchmarks for margin evaluation in basal cell carcinoma (BCC) surgery.
• An area under the receiver operating characteristic curve (AUC) of 81% was achieved.
• Attention maps were generated to assess the biological relevance of the deep learning models' findings.

Conclusions:

• A novel method for characterizing surgical margins using iKnife mass spectrometry data is proposed.
• The ImSpect framework enables the application of advanced deep learning models to iKnife data.
• This approach holds potential for improving real-time decision-making during cancer surgery.