Ciliafy - A Motion Assessment Framework for Primary Ciliary Dyskinesia Diagnosis

The air we breathe is not always clean.. It can carry bacteria and pathogens into our nose and lungs. Luckily, our respiratory system has tiny-hair like cells called cilia that act like little brooms to kick mucus and harmful particles out.

But what happens when these tiny brooms stop moving properly?

Mucus and bacteria can now build up, causing repetitive infections. This condition is known as Primary Ciliary Dyskinesia. It is a very complex disease to diagnose because its symptoms often resemble other respiratory diseases; Therefore, diagnosis is often delayed by 5-10 years.

My ciliafy framework analyzes ciliary motion using high-speed videos and computer vision. I do not just analyze how fast the cilia move, I analyze how they bend, sweep and move as a wave. By adding these novel features, my system helps detect unhealthy cilia more clearly and supports an earlier and more reliable diagnosis.

Illustration and stock videos clips used with permission from Canva and Capcut. The cilia introduction cli used with permission from Polymime

Background music used with permission from Pixabay.

Primary Ciliary Dyskinesia is an overlooked and rare genetic disorder caused by impaired ciliary motion, affecting approximately 1 in 7,500 people worldwide (Hannah et al., 2022). In a healthy system, the mucociliary escalator functions as a vital defence system (Figure 1), but in PCD, this process falls short. Resembling other respiratory disorders, it leads to diagnosis delays till adulthood; during this time many patients develop chronic lung infections and progressive respiratory complications (Collison et al., 2025).

This raised a key question: Why is this condition so difficult to diagnose accurately and early?

Current methods for analyzing the ciliary motion include Transmission Electron Microscopy (TEM) and nasal nitric oxide (nNO). However, TEM misses 30% of PCD patients who have completely "normal-looking" cilia under this method, but their cilia are functionally paralyzed (Figure 2). High-speed video microscopy (Figure 3) has emerged as an advanced method for computational analysis of ciliary motion (Rubbo et al., 2018). However, this method does not fully capture the complexity of movement, as cilia may beat at near normal frequency while exhibiting disrupted waveform geometry. One specific instance for this case is a phenotype named DNAH11, where unhealthy cilia move at frequencies similar to or greater than healthy cilia (Raidt et al., 2014).

These diagnostic gaps exist because frequency alone cannot capture the kinetic defects and disruptions in waveform geometry (Figure 4,5). This framework aims to provide an effective approach for characterization of ciliary motion and classification of abnormal beating patterns associated with PCD.

I conducted research into the inner biology of cilia, which helped me understand impaired, abnormal motion. Further exploration was aimed at specific variants of PCD, understanding how it can affect the internal structure, and the motion of cilia. It became evident that current software aimed at diagnosis rely only on speed of cilia (frequency), missing the subtle abnormalities seen in variants of PCD, as highlighted in figure 6. The development of the framework incorporated two novel features, curvature and bending angle.

The system extracts descriptive features of ciliary motion more than 900 ex-vivo high-speed videos were obtained from open-source databases. These videos were cleaned to implement the focus on active cilia, isolating them from the background and measuring the change in colour of pixels, and incorporating Fast-Fourier Transform (FFT) to analyze peak frequency in Hz (Figure 7). For bending angle, the base of cilia was treated as a fixed point (P0), and tips were detected across frames to measure the distance from base to the tips (Figure 8). The bending angle was calculated using the law of cosines. To quantify how much cilia bend, novel physics models which treat the cilium as a flexible rod were used. The cilium was treated as a rod, with its bending described along its middle line over time (Figure 9).

After these features were extracted, a machine learning pipeline was developed for binary classification. The data was split 80% for training, and 20% for testing. To evaluate their accuracy, models such as Support Vector Machine (SVM), Random Forest (RF), and XGboost were trained and tested against the clinical ground truth. These steps are fully demonstrated in the flowchart (Figure 10).

Ciliafy was developed as an automated framework to quantitatively characterize ciliary motion and identify abnormalities associated with PCD. By executing the automated feature extraction algorithms on more than 900 ex-vivo high-speed videos, three significant features were extracted. The comparative experimental results of the quantitative biomarkers are displayed in Figure 11.

Firstly, the extracted frequency values for healthy cilia demonstrated ≥ 11 Hz, whereas PCD samples exhibit average frequency < 11 Hz. The results matched the historical threshold for beat frequency seen with healthy and unhealthy samples (Stannard et al., 2009). Furthermore, rather than maintaining a steady, rhythmic pattern, PCD samples often demonstrated highly erratic beating (Figure 12). While this feature proved as a baseline for diagnosis, it was the other two novel features that captured the subtle abnormalities.

Secondly, the bending angle analysis quantifies the physical sweep of the cilia. By tracking the motion of specific tips relative to their base, P0, and applying the law of cosines the framework extracted quantitative values (Figure 13) (Papon et al., 2012). Healthy cilia, beating steadily at 11Hz, also demonstrated a bending angle ≥ 45◦ – a vital biomechanical task for the effective mucociliary clearance (Sleigh et al., 1988). On the other hand, PCD samples exhibit a bending angle < 45◦, matching their stiff behaviour, thus, preventing clearance of pathogens and toxins. Therefore, this quantitative biomarker serves as a novel feature that captures the subtleties missed by relying solely on CBF.

Curvature metrics were obtained through use of novel inverse rod-models to establish the curvature and force that causes cilia to bend by tracking its position data, across its motion for successive frames. This was done by treating the cilia as a flexible, slender rod measuring its curvature at every point on cilium across successive frames. It proved that healthy ciliary motion relies on complex, fluid bending along the middleline, thus having a parabolic arch motion, as demonstrated in Figure 14.  The unhealthy samples lacked this fluidity entirely; Therefore, their stunned motion concluded that they were physically failing to execute a sweep, therefore confirming that dyskinetic cilia can be diagnosed with multi-parameter analysis rather than just relying on a simple frequency deficit.

When evaluating the classification accuracy, the Ciliary framework approach proved highly effective. It reached a classification accuracy of 87% with XGboost, 85% with SVM, and 82% with RF (Figure 15). More importantly, by capturing their erratic speed, geometry, and curvature, this framework effectively provides a reliable solution that successfully flags the physically impaired cilia that conventional diagnostic methods often miss, helping significantly reduce delayed diagnosis.

The results conclusively prove that relying alone on CBF is an insufficient biomarker (Figure 16). Thus, I learned that a more holistic approach is needed. By incorporating quantitative geometric metrics, specifically including novel inverse-rod model (Figure 17) to capture curvature, this framework captures the subtle yet critical motion patterns that overcome the limitations in motion analysis. This biological distinction is significant: a rapidly vibrating but rigid cilium cannot clear mucus, leaving patients open to chronic infections.

The vitality of this project lies in its ability to translate complex data into a structured binary classification of motion for intuitive and rapid clinical interpretation for improved decision support. It fills the limitation in current diagnostic tools as it incorporates computational analysis using computer vision and machine learning models (Figure 18), which have reported high feasibility for diagnosing PCD, with a reported sensitivity of 0.82 - 0.90, making it suitable for screening, in addition to the conventional diagnosis methods (Burns et al., 2025).

Furthermore, the ciliafy framework leverages efficient algorithms rather than relying on expensive, time-consuming assessments (Figure 19). Clinical interpretability is achieved by replacing subjective assessments with automated analysis, empowering clinics to accurately diagnose patients and prevent long-term respiratory damage (Figure 20). To implement ciliafy into a diagnostic setting as a software, it would require incorporation of non-linear features to further refine motion capture, therefore providing a high-precision screening test, in addition to current methods, to capture the subtle abnormalities that often go missed.

Future improvements will focus on adding enhanced features that capture the subtle abnormalities, thus improving the accuracy when discriminating between the two groups. These include:

Time consistency: How steadily the cilia moves over time (Figure 21)

Periodicity: How rhythmically is their beating pattern (Figure 22)

Synchronization: How well they work together as a team (Figure 23)

Sweep Amplitude: How consistently their motion bending motion stays (Figure 24)

Also, I plan to integrate a dashboard that displays extracted features and classification results into a user-friendly application to improve decision-making and enhance diagnosis (Figure 25).

I had two respected mentors who guided me throughout the development of my project.

For the life science aspect, I sought guidance from Dr.Sachin Goyal, Associate Professor and Department Vice chair of Graduate Studies at UC Merced. Through our weekly meetings, I gained deeper knowledge of the inner biology of cilia and discussed the feasibility of deploying inverse-rod models to measure the bending patterns of cilia.

The technical refinement came from Prof. Simarjeet Saini, University of Waterloo. His expertise in algorithms and machine learning helped me improve the classification accuracy.

I am also grateful to my family members for providing the support and equipment required to conduct this research. I also thank the WWSEF team for their continuous efforts

Their collective effort is what enabled me to fill the gap for diagnosis of PCD, transforming a theoretical concept into a diagnostic tool to make a dent in the universe.

References

Arata, M., Usami, F. M., & Fujimori, T. (2022, November 11). Coordination of cilia movements in multi-ciliated cells. Journal of developmental biology. https://pmc.ncbi.nlm.nih.gov/articles/PMC9680496/

Burns, G., Kauffman, C., Manion, M., Pai, R. A., Milla, C., O'Connor, M. G., Shapiro, A. J., & Bjornson-Pennell, H. (2025). Feasibility of machine learning analysis for the identification of patients with possible primary ciliary dyskinesia. Orphanet journal of rare diseases, 20(1), 516. https://doi.org/10.1186/s13023-025-03966-z

Castillo, M., Freire, E., & Romero, V. I. (2023). Primary ciliary dyskinesia diagnosis and management and its implications in America: a mini review. Frontiers in Pediatrics, 11, 1091173. https://doi.org/10.3389/fped.2023.1091173

Chioccioli, M., Feriani, L., Nguyen, Q., Kotar, J., Dell, S., Mennella, V., Amirav, I., & Cicuta, P. (2019). Quantitative high-speed video profiling discriminates between DNAH11 and HYDIN variants of primary ciliary dyskinesia. American Journal of Respiratory and Critical Care Medicine, 199, 1432–1434.

https://doi.org/10.1164/rccm.201812-2256LE

Collison, R., Hyatali, S. A., Kamenova, A., Rashed, A., Riley, D., Kumar, K., Stowell, J. M., & Loebinger, M. R. (2025). Primary ciliary dyskinesia: Aetiology, diagnosis and clinical management. Clinical Medicine, 25(3), 100319. https://doi.org/10.1016/j.clinme.2025.100319

Ding, Y., Nawroth, J. C., McFall-Ngai, M. J., & Kanso, E. (2014). Mixing and transport by ciliary carpets: a numerical study. Journal of Fluid Mechanics, 743, 124–140. https://doi.org//10.1017/jfm.2014.36

Hannah, W. B., Seifert, B. A., Truty, R., Zariwala, M. A., Ameel, K., Zhao, Y., Nykamp, K., & Gaston, B. (2022). The global prevalence and ethnic heterogeneity of primary ciliary dyskinesia gene variants: a genetic database analysis. The Lancet. Respiratory Medicine, 10(5), 459–468.  https://doi.org/10.1016/S2213-2600(21)00453-7

Knowles, M. R., Daniels, L. A., Davis, S. D., Zariwala, M. A., & Leigh, M. W. (2013). Primary ciliary dyskinesia. Recent advances in diagnostics, genetics, and characterization of clinical disease. American  Journal of Respiratory Critical Care Medicine, 188(8), 913–922. https://doi.org/10.1164/rccm.201301-0059CI

Laligné, C., Klotz, C., Garreau de Loubresse, N., Lemullois, M., Hori, M., Laurent, F. X., Papon, J. F., Louis, B., Cohen, J., & Koll, F. (2010). Bug22p, a conserved centrosomal/ciliary protein also present in higher plants, is required for an effective ciliary stroke in Paramecium. Eukaryotic Cell, 9(4), 645–655. https://doi.org/10.1128/EC.00368-09

Lucas, J. S., Barbato, A., Collins, S. A., Goutaki, M., Behan, L., Caudri, D., et al. (2017). European Respiratory Society

guidelines for the diagnosis of primary ciliary dyskinesia. European Respiratory Journal, 49(1), 1601090. https://doi.org/10.1183/13993003.01090-2016

Mizuno, K., Shiba, K., Yaguchi, J., Shibata, D., Yaguchi, S., Prulière, G., Chenevert, J., & Inaba, K. (2017). Calaxin establishes basal body orientation and coordinates movement of monocilia in sea urchin embryos. Scientific Reports, 7, Article 10751. https://doi.org/10.1038/s41598-017-10822-z

Oldenburg, A., Chhetri, R., Hill, D., & Button, B. (2012). Monitoring airway mucus flow and ciliary activity with optical coherence tomography. Biomedical Optics Express, 3(8), 1978–1992. https://doi.org/10.1364/BOE.3.001978

Papon, J. F., Bassinet, L., Cariou-Patron, G., Zerah-Lancner, F., Vojtek, A. M., Blanchon, S., Crestani, B., Amselem, S.,Coste, A., Housset, B., Escudier, E., & Louis, B. (2012). Quantitative analysis of ciliary beating in primary ciliary dyskinesia: a pilot study. Orphanet journal of rare diseases, 7,78. https://doi.org/10.1186/1750-1172-7-78

Raidt, J., Wallmeier, J., Hjeij, R., Onnebrink, J. G., Pennekamp, P., Loges, N. T., Olbrich, H., Häffner, K., Dougherty, G. W., Omran, H., & Werner, C. (2014). Ciliary beat pattern and frequency in genetic variants of primary ciliary dyskinesia. European Respiratory Journal, 44(6), 1579–1588. https://doi.org/10.1183/09031936.00052014

Rubbo, B., Shoemark, A., Jackson, C. L., Hirst, R., Thompson, J., Hayes, J., Frost, E., Copeland, F., Hogg, C., O’Callaghan, C., Reading, I., & Lucas, J. S. (2019). Accuracy of high-speed video analysis to diagnose primary ciliary dyskinesia. Chest, 155(5), 1008–1017. https://doi.org/10.1016/j.chest.2019.01.036

Sampaio, P., da Silva, M. F., Vale, I., Roxo-Rosa, M., Pinto, A., Constant, C., Pereira, L., Quintão, C. M., & Lopes, S. S. (2021). CiliarMove: New software for evaluating ciliary beat frequency helps find novel mutations by a Portuguese multidisciplinary team on primary ciliary dyskinesia. ERJ Open Research, 7(1), 00792-2020. https://doi.org/10.1183/23120541.00792-2020

Savas, S. T., Tschanz, S. A., Latzin, P., Casaulta, C., & Müller, L. (2024, January 1). Similar effect of lidocaine and saline on ciliary beating of nasal epithelial cells in vitro. medRxiv.

https://www.medrxiv.org/content/10.1101/2024.03.25.24304837v1.full

Schneiter, M., Tschanz, S. A., Escher, A., Müller, L., & Frenz, M. (2023). The Cilialyzer - A freely available open-source software for the analysis of mucociliary activity in respiratory cells. Computer Methods and Programs iBiomedicine, 241, 107744. https://doi.org/10.1016/j.cmpb.2023.107744

Shapiro, A. J., et al. (2024). Asthma diagnosis in children with primary ciliary dyskinesia. JAMA Network Open, 7(11), e2827617. https://doi.org// 10.1001/jamanetworkopen.2024.27617

Sleigh, M. A., Blake, J. R., & Liron, N. (1988). The propulsion of mucus by cilia. American Review of Respiratory Disease, 137(3), 726–741. https://doi.org/10.1164/ajrccm/137.3.726

Sourav Pan. (2024, March 21). Flagella Lecture Video Definition, Structure, Types, Functions, Rotation, Examples. Biologynotesonline.com. https://biologynotesonline.com/locomotion-in-protista/