3.1.7. Hematological Cancer

Leukemic diseases are a diverse group of neoplasms that result from genetic disorders affecting hematopoietic precursor cells, and they represent one of the most common forms of hematologic cancer globally. Accurate diagnosis of these disorders requires specialized expertise and often involves using multiple techniques [98]. In this section, we presented the use of optical biosensors in conjunction with machine learning for detecting hematological cancer.

DNA methylation is a process in which a methyl group is attached to the fifth carbon atom of a cytosine (C) residue, resulting in the formation of 5-methylcytosine (5-mC). The methylation patterns in cancer genomes exhibit unique characteristics, known as the methyl cape, and can serve as a potential universal biomarker for cancer detection [99]. For example, Koowattanasuchat et al. [99] presented the development of a methyl cape sensing platform for leukemia screening using cysteamine-decorated gold nanoparticles (Cyst/AuNPs). The platform is based on methylation-dependent DNA solvation, and normal and cancerous DNAs have distinct methylation profiles. The authors report 95.3% accuracy in leukemia screening using an optical spectrophotometer and 90% accuracy when a smartphone system is used.

Minimal residual disease (MRD) testing is used mostly for blood cancers, such as lymphoma and leukemia [100,101]. Uslu et al. [102] investigated the signal readout mechanism of a biochip designed to detect MRD, which refers to highly resistant cancer cells that can cause relapse in cancer survivors after treatment. To improve the capture, isolation, and counting of these tumor cells, the team combined previously explored methods with the use of immunomagnetic beads. These beads are coated with receptors that bind to and capture target molecules, allowing them to be manipulated using magnetic fields. Once the unbound beads were filtered out of the microfluidic channel, the remaining beads were

imaged at 20× and 40× magnifications using a CCD camera and processed using computer vision. The authors demonstrated the accuracy and reproducibility of the method through various experiments and comparisons with manual counting. They also discussed the potential applications of the automated method in research and clinical settings for the detection and monitoring of leukemia and other diseases. Machine learning algorithms to analyze the images obtained were utilized. Among the algorithms tested, the RF algorithm achieved the highest accuracy of 87.4%.

Tremendous progress has been made in the field of cancer treatment through the utilization of high-affinity T-cell receptors and chimeric antigen receptor (CAR)-modified T cells. These innovative approaches have recently obtained approval from the Food and Drug Administration (FDA) for treating certain hematologic malignancies [103]. To demonstrate, Sarkar et al. [103] implemented the droplet microfluidics-based cytotoxicity imaging approach to isolate individual natural killer cells. They measured their ability to kill cancer cells in the presence of different types of antibodies. Machine learning algorithms for analyzing the resulting data were used, and they predicted which types of antibodies were most effective in activating the natural killer cells.

Last but not least, Li et al. [104] presented a novel approach to improving the accuracy of blood cancer cells and biomarker identification in label-free flow cytometry using parallel quantitative phase imaging. Such technology holds promise for the early detection of primary cancer or metastasis. The team used this imaging technique to assess additional parameters, such as cell protein concentration, allowing for increased accuracy in categorizing unlabeled cells. Additionally, they developed a CNN that directly operated on the measurement signals of this setup to detect cancer cells more efficiently. They demonstrated the applicability of the new method in the classification of white blood cells and epithelial cancer cells with more than 95% accuracy in a label-free fashion. Table 1 provides a summary of the cancer cell types that were detected using optical biosensors, along with the outcomes of the machine learning algorithms applied to the data.


**Table 1.** Comparison of different optical-based biosensors with ML analysis for cancer cell detection.

