Using physiological parameters measured by hyperspectral imaging to detect colorectal cancer

Area of Science:

• Medical imaging
• Machine learning
• Colorectal surgery

Background:

• Accurate malignant tissue detection is crucial for colorectal surgery outcomes.
• Non-invasive spectral imaging with machine learning (ML) shows promise for tumor identification.
• Large spectral ranges in ML increase computing time, necessitating feature reduction.

Purpose of the Study:

• To evaluate ML methods for automatic classification of cancerous and healthy colon tissue.
• To reduce computational time by using specific physiological tissue parameters and wavelength bands.

Main Methods:

• Evaluated logistic regression and convolutional neural network (CNN) ML models.
• Utilized four physiological tissue parameters for classification.
• Focused on specific wavelength bands to reduce spectral range.

Main Results:

• Achieved a Receiver Operating Characteristic Area Under the Curve (ROC AUC) of 0.81 with the CNN.
• Demonstrated that specific wavelength bands can effectively detect cancer.
• Showcased the potential for automatic classification based on physiological parameters.

Conclusions:

• Specific wavelength bands combined with ML can accurately classify colon tumors.
• This approach supports real-time, image-guided colorectal surgery.
• Future applications include enhanced intraoperative decision-making in oncology.