Currently, the primary basis for diagnosis of cancer is evaluation of morphological changes in a tissue biopsy by a trained pathologist, a process with inherent subjectivity. In contrast to traditional tissue biopsies, of which collection is invasive to the patient and requires preliminary knowledge on the suspected body location, liquid biopsies, such as blood and urine, can be obtained in routine lab tests. Various strategies have been developed for detection of cancer cells in liquid biopsies, mostly relying on antibody-based capturing of cells like fluorescence-activated cell sorting and magnetic-activated cell sorting. However, for many cell types, antigen combinations that would allow for their unambiguous identification are missing. Moreover, the inherent modification of the cell surface chemistry makes label-based approaches incompatible with non-destructive cell processing and, therefore, disqualifies them for usage in further cell therapeutic applications. Finally, these methods do not yield high enough specificity in charactering individual cancer cells in a reasonable amount of time and efforts, preventing their practical integration into clinical procedures. This critical unmet need can be addressed by using highly informative, label-free, imaging techniques that allow for non-invasive and automated cell processing, and at the same time offer high discriminative power on the level of the individual cell with reasonable cost. Most existing label-free analysis approaches are based on simple structural properties of cells such as general size and shape. However, these parameters provide only a moderate degree of specificity to cell type, and correlation to clinical condition is often weak. Specifically, in the field of bladder cancer monitoring, there is no reliable urine test that can be used for monitoring cancer progression, sentencing the patients to live-long invasive procedures. To solve this, we developed a new device for analysing cancer cells in liquid biopsies during flow and adapt it to analysis of cancer cells in urine samples. The device is based on label-free interferometric imaging, with adaptive-real-time cell-classification based on a deep-learning platform. The captured interferograms encapsulate the profile of the refractive index (RI) of the entire cell structure, including its interior structure, which is related to the optical interaction of the light field with cellular organelles and their physio-chemical composition. This type of information is accessible by label-free interferometric phase microscopy (IPM) without affecting the cell physiology. The RI distribution of a cell provide unique access to the cell morphology, dry mass, density and membrane surface area, parameters that are not accessible in regular imaging. Since IPM captures ‘deep’ imaging data, which is provided to the deep-learning classifying platform, clinical-level diagnosis specificity can be obtained with relatively modest real-world dataset. IPM is based on a mature technology for wavefront sensing, digital holography. However, till recently it could not be implemented in clinical settings due to its bulkiness, non-portability, extremely high sensitivity to ambient vibration and the requirement for specific optical skills to routinely align and operate it. In recent years, we made significant efforts to make these wavefront sensing configurations affordable to clinical use [9-11]. Our approach is based on attaching a portable interferometric module to the exit port of the imaging system, instead of building the interferometer around the sample, which is instable and hard to align. The resulting wavefront sensor is compact, inexpensive, mechanically stable, easy to operate, and can be attached to existing imaging systems, making this technology accessible and affordable to clinics’ direct use. In the proposed PoC project, we focused on detecting cancer cells in urine samples of bladder-cancer patients, as a means of avoiding frequent painful cystoscopies, the current standard-of-care test when monitor cancer recurrence.
