In tissue physiology, the chemical information from metabolomics provides a means of elaborating tumor characteristics at cellular level. However, obtaining such information requires invasive procedures which can be costly, delaying clinical patient management.
Scanning for Medical Diagnosis
Computed tomography (CT) is a clinical standard of care used for cancer patients. However, this imaging technique does not provide histological or prognostic information. Additionally, the ability to embed metabolomic data into CT to use the learned representation for prognosis has yet to be described.
Several studies have developed imaging analysis channels to assess and diagnose lung cancers. Techniques such as radiomics have helped extract features from CT scans, followed by the use of machine learning models.
These methods effectively quantify the spatial complexity of the tumors, like size, shape, and intensity. Just recently, experts have attempted to replace radiomic features with deep learning features taken from convolutional neural networks (CNN) directly on lesions.
In terms of medical diagnosis from CT scans, deep learning methods seem to outperform traditional radiomics extraction and selection. In practice, however, such features have limited clinical performance.
Virtual Lung Cancer Biopsy
At Imperial College London, researchers have developed a technique which involves the use of artificial intelligence in extracting information regarding the chemical composition of lung tumors from medical scans. Their study is discussed in the paper "AI Combined With Medical Imaging To Extract Information About Lung Tumors Provides Virtual Biopsy for Cancer Patients".
Led by Marc Buobnovski Martell, the research team used a non-invasive method to classify the type of lung cancer and to predict its likelihood of progression. Diagnosis is very crucial in selecting the right treatment for a disease, and the scientists believe that their technique allows doctors to do so without obtaining a physical tissue biopsy.
Recently, AI has been used in analyzing medical scans and in looking for signs of disease that are not very obvious. Generative AI is currently being utilized for multiple applications since it has the ability to learn from data to create new content.
Using these capabilities, the Imperial team wondered if medical information about lung tumors contained in the metabolomic profile can be revealed by CT scans. In order to find out, they used data from 48 lung cancer patients who received treatment at the University Hospital Reina Sofia (UHRS) in Córdoba, Spain.
All the patients underwent CT scans and detailed metabolomic profiling of their tumor tissue as well as healthy tissue near the tumor. The data gathered was used to develop an AI-powered assessment tool known as tissue-metabolomic-radiomic-CT (TMR-CT).
Martell and his colleagues discovered a significant and powerful correlation between the metabolomic profiles and deep features of the patients' CT scans. Such correlation appeared as brighter or darker areas in the scanning image. This has led the scientists to conclude that it is possible to bypass the need to obtain physical tissue samples, and gather tumor metabolic characteristics from the CT scan alone.
To further test the effectiveness of their method, the research team used their TMR-CT model in a separate group of lung cancer patients who received treatments from other hospitals. All of them had a CT scan, but there was no available metabolomics data. The results reveal that TMR-CT expertly classified lung cancer and provided dependable predictions about patient outcomes. It even surpassed the performance of traditional methods and clinical assessments based on CT.
Check out more news and information on Biopsy in Science Times.
