# Machine Learning Characterization of Cancer Patients-Derived Extracellular Vesicles using Vibrational Spectroscopies

**Authors:** Abicumaran Uthamacumaran, Samir Elouatik, Mohamed Abdouh, Michael Berteau-Rainville, Zu-hua Gao, Goffredo Arena  
**arXiv:** [2107.10332](https://arxiv.org/abs/2107.10332)  
**Date:** 2021-07-21  
**Summary page:** https://www.summarizepaper.com/en/paper/2107.10332v9/

## Key points

- Pilot study explored potential of vibrational spectroscopies in early cancer detection
- Analyzed extracellular vesicles (EVs) from blood sera of patients with different cancer subtypes and healthy controls
- Used Raman spectroscopy for EVs from 9 patients and FTIR spectroscopy for 2 cancer subtypes
- Machine learning algorithms (AdaBoost Random Forest Classifier, Decision Trees, SVM) used for classification
- ML classifiers accurately distinguished between cancer patient EVs and healthy controls with over 90% accuracy in specific spectral frequency range
- FTIR spectroscopy demonstrated an 80% classification accuracy for distinguishing between cancer EVs and healthy controls
- Study highlights potential of using machine learning algorithms with vibrational spectroscopy for liquid biopsy-based early cancer screening
- Emphasizes importance of interdisciplinary approaches in precision healthcare and computational systems medicine

## Summary

In a pilot study conducted by a team of researchers from various institutions including Concordia University, Université de Montréal, McGill University Health Centre, and others, the potential of vibrational spectroscopies in early cancer detection was explored. The study focused on analyzing extracellular vesicles (EVs) extracted from the blood sera of patients with different cancer subtypes (colorectal cancer, hepatocellular carcinoma, breast cancer, and pancreatic cancer) as well as healthy controls. Multiple Raman spectroscopy measurements were performed on EVs from 9 patients, while Fourier Transform Infrared (FTIR) spectroscopy measurements were conducted on two of the four cancer subtypes. The data obtained from these analyses were then subjected to machine learning algorithms including AdaBoost Random Forest Classifier, Decision Trees, and Support Vector Machines (SVM) for classification purposes. The results showed that the ML classifiers could accurately distinguish between the Raman spectra of cancer patient EVs and those of healthy controls with a classification accuracy exceeding 90% when focusing on a specific spectral frequency range. FTIR spectroscopy also demonstrated an 80% classification accuracy for distinguishing between cancer EVs and healthy controls. The findings highlight the potential of using basic machine learning algorithms in conjunction with vibrational spectroscopy techniques for liquid biopsy-based early cancer screening. The study emphasizes the importance of interdisciplinary approaches in improving precision healthcare and computational systems medicine. Additionally, the research team disclosed no conflicts of interest and shared their data availability statement along with funding details supporting their work. Overall, this pilot study showcases the promising role of vibrational spectroscopies and machine learning in advancing early cancer detection methods.