The Need for Real-Time Molecular Diagnosis

During cancer surgery, determining whether tissue margins are clear of cancer cells is critical for patient outcomes, but current methods create significant challenges. Surgeons routinely send tissue samples for frozen section analysis to assess surgical margins while patients remain under anesthesia, extending procedure time by 20-30 minutes per sample. However, frozen section pathology achieves only 85-90% concordance with permanent section gold standard diagnosis, creating uncertainty about margin status that can lead to either incomplete tumor removal or unnecessary tissue loss. Fluorescence-guided surgery provides valuable spatial localization of tumors but lacks the molecular specificity needed for definitive cancer diagnosis, particularly at infiltrative margins where cancer cells are sparse. Our AI-enhanced Raman spectroscopy addresses these limitations by providing high accuracy diagnostics real-time and in situ, eliminating the need for tissue removal and pathology delays while offering the molecular precision that complements fluorescence localization—creating a comprehensive system where wide-field fluorescence identifies suspicious areas and Raman spectroscopy provides definitive chemical diagnosis.

---

How AI-Raman Diagnosis Works: Raman diagnosis works through a three step process during operations. Laser Spectroscopy: A focused laser beam illuminates tissue samples, causing molecules to vibrate and scatter light at characteristic frequencies that create unique "molecular fingerprints" for different tissue types. Tissue Analysis: The Raman spectrometer captures these scattered light signals, measuring the intensity and frequency shifts that correspond to specific molecular components like proteins, lipids, and nucleic acids present in cancerous versus normal tissue. AI Classification: Machine learning algorithms trained on thousands of tissue spectra analyze these molecular fingerprints in real-time, comparing new samples against validated cancer signatures to provide instant diagnostic classification with high accuracy. This system converts invisible molecular vibrations into definitive cancer detection within seconds of tissue contact.

---

Our Research Areas:
Deep Learning Algorithm Development 1D and 2D convolutional neural networks that analyze Raman spectral data for cancer classification, achieving over 99% accuracy in distinguishing tumor from normal tissue.
Multi-Cancer Detection Systems Spectroscopic approaches that identify not only the presence of cancer but also determine the organ of origin for metastatic tumors, supporting staging and treatment planning.
Real-Time Clinical Integration Handheld Raman devices and processing systems designed for intraoperative use, providing instant molecular analysis during surgical procedures.

---

Featured Publications:
"Accurate Categorization and Rapid Pathological Diagnosis Correction with Micro-Raman Technique in Human Lung Adenocarcinoma Infiltration Level" - Translational Lung Cancer Research (2024) Clinical validation study using 19,000 Raman spectra from 59 lung adenocarcinoma patients, demonstrating 99.0% AUC in detecting different levels of tumor infiltration using 1D-CNN models. DOI:10.21037/tlcr-24-168