Noninvasive detection for bladder cancer: Quantitative interferometric imaging flow cytometry

Abstract

Noninvasive detection is crucial for achieving a convenient and painless diagnosis of bladder cancer. In a recent report published in Cytometry Part A, Matan Dudaie and coworkers have successfully employed a combination of quantitative interference imaging flow cytometry and machine learning to achieve a noninvasive, label-free approach for detecting bladder cancer cells in urine samples [1].

Noninvasive detection of bladder cancer based on urine samples has been a highly challenging problem. So far, the gold standard for clinical diagnosis still relies on invasive methods such as cystoscopy and tissue biopsy [2]. These approaches not only have high costs but also carry a certain risk of infection and other side effects after testing, greatly increasing the burden on patients. As the excreted substance of the bladder, urine is the most valuable detection medium [3]. In clinical practice, urine cytology is sometimes used to screen for bladder cancer cells, but the efficiency of this method is very low. To achieve a fast and noninvasive detection method, people have also tried to use flow cytometry to detect urine samples [4]. However, the scattering parameters of flow cytometry are insufficient to differentiate between bladder cancer and normal cells, and relying on fluorescence intensity poses a heightened risk of false positives. These reasons have hindered the effective development of noninvasive detection methods for bladder cancer. Therefore, the development of noninvasive methods for detecting bladder cancer is crucial for reducing the burden on patients.

Matan Dudaie and coworkers presented their efforts in developing a novel imaging flow cytometry method for noninvasive detection of bladder cancer in Cytometry Part A. By constructing a quantitative interferometric imaging flow cytometry system, they achieved label-free detection of bladder cancer. Their detection unit consists of microfluidic channels and a quantitative interferometric microscope, and the image processing unit is composed of a convolutional neural network (CNN). The key advantage lies in achieving noninvasive, label-free, highly accurate detection of bladder cancer cells simply by collecting urine samples.

Imaging flow cytometry, as a novel method for cell analysis, can be considered a fusion of optical microscopy and flow cytometry. It enables high-throughput and high-content cell imaging, thereby enhancing the efficiency of morphology-based cell analysis. Currently, the commonly used imaging flow cytometry technique collects images based on intensity imaging principles, where intensity often represents cell morphology but struggles to convey information about the cellular metabolic state.

The refractive index, as an intrinsic optical property of cells, can provide information about the cellular metabolic state [5]. Through quantitative phase imaging technology, refractive index information of cells can be obtained. Research has indicated that refractive index-related features can serve as biomarkers for cancer diagnosis. Matan Dudaie and coworkers combined quantitative phase imaging with imaging flow cytometry to achieve high-throughput capture of cellular phase images. They validated the system's feasibility through in vitro experiments, utilizing a mixed-cell model comprising eight different stages of bladder cell, normal epithelial cells, white blood cells, and red blood cells for high-sensitivity phase imaging. Semantic segmentation CNNs were employed to obtain individual cell phase images through image cropping. By constructing two neural networks (QIFC-DL–Mobilenet-V2 and QIFC-XGB–Extreme gradient boosting [XGB] algorithm) based on image and feature approaches, they achieved high accuracy in cell classification.

To validate the clinical feasibility, Dudaie and colleagues performed quantitative phase imaging on 164,775 cells from the urine of bladder cancer patients and healthy volunteers. Through feature analysis of the two sample groups, the results revealed significant differences in characteristics between the two patient cohorts, once again confirming the ability of phase imaging to capture differences between bladder cancer cells and normal cells. By employing two types of neural networks developed in vitro, high-accuracy predictions were achieved.

This novel bladder cancer detection method eliminates the need for invasive tissue penetration, staining, and manual identification, achieving high accuracy in bladder cancer detection solely through urine analysis. Therefore, this approach holds promise as a new method for early screening of bladder cancer. For future advancement, this technology has the potential to achieve more comprehensive training of the network by gathering a more diverse range of samples. Consequently, it can also achieve high accuracy in bladder cancer detection when faced with complex clinical diagnostic environments.

Regarding the last issue, there are currently several approaches available for reference. Cell sorting, as a relatively mature method for cell purification, would be appropriate for extracting high-purity cancer cells as a training set [6]. A Field-Programmable Gate Array (FPGA) is a flexible and programmable integrated circuit capable of real-time processing and analyzing collected cell data, allowing for the selective capture of cell images [7, 8]. Interpreting high-content image information for bladder cancer diagnosis remains challenging, but this problem can be addressed by establishing meaningful features through the correlation of biochemical analysis and image characteristics.

Shubin Wei: Conceptualization; writing – original draft. Cheng Lei: Conceptualization; writing – original draft; writing – review and editing.

The authors declare no conflicts of interest.