NON-INVASIVE DIAGNOSIS OF CHRONIC KIDNEY DISEASE BY MACHINE-LEARNING BASED EVALUATION OF URINARY EXFOLIATED PROXIMAL TUBULE CELL MULTISPECTRAL AUTOFLUORESCENCE

Introduction:

Early diagnosis of CKD is important to ensure individuals receive timely intervention if required, to optimize clinical outcomes and reduce risk of progression to end-stage kidney disease. Kidney biopsy remains the gold-standard option to diagnose CKD. However, routine kidney biopsy is costly and is associated with pain and bleeding. Given such concerns, there is a critical need to explore alternative non-invasive diagnostic options in CKD. Kidney cell exfoliation into urine is an active process, which may generate significant information regarding an individual's kidney status. In particular, proximal tubule cells (PTCs) make up the majority of the kidney mass and represent the hallmark of CKD in reflecting tubulointerstitial fibrosis and atrophy. PTC autofluorescence is known to be sensitive to metabolic changes and oxidative stress, key mechanisms linked to kidney function decline. We evaluated the innovative use of multispectral autofluorescence imaging on urinary exfoliated PTCs as a non-invasive diagnostic option in CKD, where cell autofluorescence features between individuals with varying levels of kidney function (as determined by eGFR) were compared via a machine learning-based approach. 

Methods:

Individuals aged 18 or above were included. Spot urine was collected and stored at -80°C. On assessment, urinary exfoliated PTCs were specifically extracted using an immunomagnetic separation method based on anti-CD13 and anti-SGLT2 antibodies. Imaging was performed on a customized multispectral Olympus IX83 microscope with a cooled, low-noise Nüvü™ EMCCD camera and a multi-LED light source. For each individual sample, 34 combinations of excitation and emission wavelengths were acquired. Image analysis involved data preprocessing and then spectral feature analysis, performed through linear support vector machine-learning classification using an original MATLAB-based graphical user interface software developed by our group. Three discriminative studies were conducted - an overall analysis between exfoliated PTCs of individuals with eGFR≥60 and eGFR<60 plus two subgroup analysis, one between that of individuals with eGFR≥90 and eGFR 60-90 and the other between those with eGFR 60-90 and eGFR<60. 5-fold cross validation was applied in generating a mean AUC value. 

Results:

60 individuals were involved. 40 individuals had eGFR≥60 and 20 had eGFR<60. Assessment of cell autofluorescence features differentiated between exfoliated PTCs of these 2 groups with an AUC value of 0.81±0.04 (Figure 1). In the subgroup analysis, 20 individuals had eGFR≥90 and 20 had eGFR 60-90. Assessment of cell autofluorescence features differentiated between exfoliated PTCs of these 2 groups with an AUC value of 0.74±0.05 (Figure 2). Assessment of cell autofluorescence features differentiating between exfoliated PTCs from the 20 individuals with eGFR 60-90 and the 20 with eGFR<60 displayed an AUC value of 0.80±0.08 (Figure 3).

Conclusions:

Our findings from this novel approach demonstrate a good degree of differentiation between urinary exfoliated PTCs derived from individuals across varying levels of kidney function – between those with normal or high eGFR, early stages of CKD, and those with CKD. Multispectral autofluorescence imaging of urinary exfoliated PTCs combined with machine learning-based spectral feature analysis may potentially be an effective non-invasive option for CKD diagnosis. 

I have no potential conflict of interest to disclose.

I did not use generative AI and AI-assisted technologies in the writing process.