Search for: All records

The development of noninvasive methods for bladder cancer identification remains a critical clinical need. Recent studies have shown that atomic force microscopy (AFM), combined with pattern recognition machine learning, can detect bladder cancer by analyzing cells extracted from urine. However, these promising findings were limited by a relatively small patient cohort, resulting in modest statistical significance. In this study, we corroborated the AFM technique’s capability to identify bladder cancer cells with high accuracy using a controlled model system of genetically purified human bladder epithelial cell lines, comparing cancerous cells with nonmalignant controls. By processing AFM adhesion maps through machine learning algorithms, following previously established methods, we achieved an area under the ROC curve (AUC) of 0.97, with 91% accuracy in cancer cell identification. Furthermore, we enhanced cancer detection by incorporating multiple imaging channels recorded with AFM operating in Ringing mode, achieving an AUC of 0.99 and 93% accuracy. These results demonstrated strong statistical significance (p < 0.0001) in this well-defined model system. While this controlled study does not capture the biological variation present in clinical settings, it provides independent support for AFM-based detection methods and establishes a rigorous technical foundation for further clinical development of AFM imaging-based methods for bladder cancer detection. 

Here, we present a novel mechano-spectroscopic atomic force microscopy (AFM-MS) technique that overcomes the limitations of current spectroscopic methods by combining the high-resolution imaging capabilities of AFM with machine learning (ML) classification. AFM-MS employs AFM operating in sub-resonance tapping imaging mode, which enables the collection of multiple physical and mechanical property maps of a sample with sub-nanometer lateral resolution in a highly repeatable manner. By comparing these properties to a database of known materials, the technique identifies the location of constituent materials at each image pixel with the assistance of ML algorithms. We demonstrate AFM-MS on various material mixtures, achieving an unprecedented lateral spectroscopic resolution of 1.6 nm. This powerful approach opens new avenues for nanoscale material study, including the material identification and correlation of nanostructure with macroscopic material properties. The ability to map material composition with such high resolution will significantly advance the understanding and design of complex, nanostructured materials. 

Steps of combining atomic force microscopy (AFM or SPM) with machine learning. 

Previously, the analysis of atomic force microscopy (AFM) images allowed us to distinguish normal from cancerous/precancerous human epithelial cervical cells using only the fractal dimension parameter. High-resolution maps of adhesion between the AFM probe and the cell surface were used in that study. However, the separation of cancerous and precancerous cells was rather poor (the area under the curve (AUC) was only 0.79, whereas the accuracy, sensitivity, and specificity were 74%, 58%, and 84%, respectively). At the same time, the separation between premalignant and malignant cells is the most significant from a clinical point of view. Here, we show that the introduction of machine learning methods for the analysis of adhesion maps allows us to distinguish precancerous and cancerous cervical cells with rather good precision (AUC, accuracy, sensitivity, and specificity are 0.93, 83%, 92%, and 78%, respectively). Substantial improvement in sensitivity is significant because of the unmet need in clinical practice to improve the screening of cervical cancer (a relatively low specificity can be compensated by combining this approach with other currently existing screening methods). The random forest decision tree algorithm was utilized in this study. The analysis was carried out using the data of six precancerous primary cell lines and six cancerous primary cell lines, each derived from different humans. The robustness of the classification was verified using K-fold cross-validation (K = 500). The results are statistically significant at p < 0.0001. Statistical significance was determined using the random shuffle method as a control. 

Here we address an important roadblock that prevents the use of bright fluorescent nanoparticles as individual ratiometric sensors: the possible variation of fluorescence spectra between individual nanoparticles. Ratiometric measurements using florescent dyes have shown their utility in measuring the spatial distribution of temperature, acidity, and concentration of various ions. However, the dyes have a serious limitation in their use as sensors; namely, their fluorescent spectra can change due to interactions with the surrounding dye. Encapsulation of the d, e in a porous material can solve this issue. Recently, we demonstrated the use of ultrabright nanoporous silica nanoparticles (UNSNP) to measure temperature and acidity. The particles have at least two kinds of encapsulated dyes. Ultrahigh brightness of the particles allows measuring of the signal of interest at the single particle level. However, it raises the problem of spectral variation between particles, which is impossible to control at the nanoscale. Here, we study spectral variations between the UNSNP which have two different encapsulated dyes: rhodamine R6G and RB. The dyes can be used to measure temperature. We synthesized these particles using three different ratios of the dyes. We measured the spectra of individual nanoparticles and compared them with simulations. We observed a rather small variation of fluorescence spectra between individual UNSNP, and the spectra were in very good agreement with the results of our simulations. Thus, one can conclude that individual UNSNP can be used as effective ratiometric sensors. 

It has been recently demonstrated that atomic force microscopy (AFM) allows for the rather precise identification of malignancy in bladder and cervical cells. Furthermore, an example of human colorectal epithelial cells imaged in AFM Ringing mode has demonstrated the ability to distinguish cells with varying cancer aggressiveness with the help of machine learning (ML). The previously used ML methods analyzed the entire cell image. The problem with such an approach is the lack of information about which features of the cell surface are associated with a high degree of aggressiveness of the cells. Here we suggest a machine-learning approach to overcome this problem. Our approach identifies specific geometrical regions on the cell surface that are critical for classifying cells as highly or lowly aggressive. Such localization gives a path to colocalize the newly identified features with possible clustering of specific molecules identified via standard bio-fluorescence imaging. The biological interpretation of the obtained information is discussed.