Algorithm based analysis of circulating tumour cells in cancer patients

Introduction: 

The role of circulating tumour cells (CTCs) in solid cancers for prognostication, treatment response, and early metastasis detection is well established. Based on fluorescence microscopy and fluorescence-activated cell sorting (FACS), the enumeration of CTCs with varied protein biomarkers has routinely been performed. CTC detection methods require more robust image analysis. We developed CTC image analysis using an algorithm that assesses red, green, and blue fluorescent signals along with cellular morphological traits. We compared the concordance and outcome with manual image analysis.

Methods: 

Retrospectively, we processed blood samples from 32 cancer patients (27 at baseline and 5 follow-ups) to determine CTC prevalence. Cancer types included colorectal (n=16), breast (n=9), glioblastoma (n=2), periampullary (n=1), urothelial (n=1), ovary (n=1), lung (n=1), and melanoma (n=1). To capture CTCs, we processed 1.5 mL of peripheral blood using the OncoMetastat device, with affinity-conjugated glass beads (GBs). CTC images were captured and confirmed on GBs based on CK18-positive (Green), DAPI-positive (Blue), and CD45-negative (Red) signals using a Zeiss Microscope (Method 1). Further, the cells were qualified using brightfield imaging and cell size. Cells were selected based on the highest CTC traits, namely size, shape, and intensities. Furthermore, we developed a Python-based image analysis algorithm, “CellMatrics,” for CTCs based on morphology, size, brightfield, and RGB intensities (Method 2). The CTC images were imported into the algorithm using a drag-and-drop interface. The algorithm applies broad criteria for RGB intensities and finally assesses true positives while eliminating false negatives in 10 seconds. The outcome is represented by the CTC image, count, RGB intensities, size, and surface area. Finally, we compared the concordance between CTCs identified manually and those identified by the algorithm.

Results: 

A total of 24 CTCs were detected in 23/32 (71.88%) cancer patients using Method 1. The number of CTCs ranged from 1–3, and the overall mean CTC distribution was approximately 1.04. Patients aged 20–29, 30–39, and 40–49 years demonstrated the highest mean CTC count (1.00), while the lowest mean CTC count (0.50) was observed in patients aged 50–59 years. Colorectal samples showed 11 CTCs, breast showed 6, ovary showed 3, and samples from periampullary, urothelial, glioblastoma, and lung cancer types each showed 1 CTC. The melanoma sample showed 0 CTCs. Upon importing CTC images into the algorithm (Method 2), an identical CTC outcome was observed.

Conclusion: 

We evaluated CTCs in blood samples from 32 cancer patients. Comparing the outcome from manual CTC analysis (Method 1) with Method 2 qualified the CTCs detected using the automated algorithm. We observed rapid and robust concordance in CTC counts, matching the manual CTC count.