Metabolism Imaging-genomics for Predicting the Surgical Outcomes of Colorectal Cancer

Overview

In this study, the investigators constructed an imaging-metabolism prediction model for colorectal cancer by analysing the imaging and metabolomics features of colorectal cancer, in order to further adjust and guide the treatment plan.

This experiment is a prospective cohort study and is expected to include 300 patients who were diagnosed with colorectal cancer and underwent radical colorectal cancer surgery at the First Affiliated Hospital of Chongqing Medical University. 1\. Data collection before and after surgical treatment 1. Study subjects: patients diagnosed with colorectal cancer and voluntarily undergoing radical colorectal cancer surgery in the First Affiliated Hospital of Chongqing Medical University will be included; preoperative imaging images will be collected before surgery, and 2 ml of blood specimens will be collected from the patients for imaging and metabolomics studies. 2. Follow-up efficacy judgement: the study subjects included in this project were followed up after undergoing surgery, and the overall survival (OS) and disease-free survival (DFS) of the study subjects should be followed up for at least 3 years. 2\. Using pre-treatment colorectal cancer imaging images, study the imaging histological features that predict the prognosis of colorectal cancer patients, and develop an imaging histological prediction model for colorectal cancer. 1. Construct an imaging histology prediction model for colorectal cancer based on the imaging images of the study subjects, and then randomly assign them to the training cohort and validation cohort by the computer according to 7:3 to carry out model evaluation. For imaging histology model development and internal validation, deep learning methods are applied for imaging histology feature extraction to extract two main categories of image features: one is manually defined features such as tumour shape, intensity, texture and wavelets, and the other is non-specified features extracted using convolutional neural networks. For the manually defined features, the investigators extract the features of colorectal cancer cancer foci. The cancer foci are outlined on the imaging images by radiologists, and four types of image features, namely, shape, intensity, texture and wavelet transform, are extracted from the cancer foci. The specific requirements are: the features extracted by shape are compatible with the doctor\'s visual judgement, e.g., the outline is not glossy, the edges are blurred, etc.; the features extracted by intensity can reflect the homogeneous nature of the foci, etc.: the features extracted by texture and wavelet transform belong to the high-dimensional complex features. can mine information that cannot be extracted by doctors\' vision. For feature extraction by convolutional neural network, the region of interest (ROI) of colorectal cancer foci is firstly sketched, and adaptive features are learnt for specific targets through supervised learning. Convolutional neural network carries out layer-by-layer nonlinear transformation and convolution operation on massive image data, and its learnt features are more targeted and adaptive than the manually designed features. Fusing the two major categories of extracted image features, the idea of combining statistical analysis and multiple machine learning feature selection methods is used to screen all the extracted features for stability and repeatability, and key features with high stability, high differentiation and high independence are obtained for subsequent development of prediction models. 2. To develop an imaging histology prediction model for colorectal cancer and elucidate the feasibility of imaging histology in predicting the prognosis of colorectal cancer patients. Using the colorectal cancer imaging features, clinical information and pathological diagnostic information selected from the research subjects included in the study as input information, and by combining a variety of existing advanced machine-learning algorithms (vector machine, random forest, AdaBoost, deep learning, etc.), cross validation is used to ensure the reliability and stability of the model, and ultimately a final design is developed that can use imaging to predict the prognosis of colorectal cancer. Imaging histology prediction model. The model should be internally validated in the validation cohort first after its establishment, and should also be externally validated because external validation of model data is generally considered to be more independent and strengthens the validation to elucidate the feasibility of imaging histology to evaluate the prognosis of colorectal cancer. 3\. Development of imaging-metabolomics prediction models related to colorectal cancer prognosis. 1. To study the metabolomic markers related to the prognosis of colorectal cancer patients using cancer tissues and blood specimens of the study subjects. 2. Metabolomics testing was performed based on blood specimens from the study subjects. Metabolomic fingerprinting was used for the metabolomics study, in which liquid chromatography-mass spectrometry (LC-MS) was used to compare the respective metabolites of treatment-resistant and effective ESCC tissues in order to identify all the metabolites therein. Metabolic fingerprinting involves comparing the mass spectrometry peaks of metabolites in different individuals to ultimately understand the structure of different compounds and to establish a complete set of analytical methods for identifying the characteristics of these different compounds. The steps of non-targeted metabolomics assay are as follows: 1) Obtaining sample information: quality control (QC) samples are prepared for determining the state of the instrument and the equilibrium chromatography-mass spectrometry (CS-MS) system prior to sampling and are used to evaluate. The stability of the system throughout the experiment.2) Sample pre-processing.3) Chromatography-mass spectrometry analysis.4) Data processing to identify differential metabolites.5) Differential metabolite metabolism.6) Sample analysis.

Conditions

Interventions

Eligibility

Locations (1)

Outcomes