Leveraging mid-infrared spectroscopic imaging and deep learning for tissue subtype classification in ovarian cancer

Abstract

Mid-infrared spectroscopic imaging (MIRSI) is an emerging class of label-free techniques being lever- aged for digital histopathology. Modern histopathologic identification of ovarian cancer involves tissue staining followed by morphological pattern...

Cervical Cancer Tissue Analysis Using Photothermal Midinfrared Spectroscopic Imaging

“…This increase is attributed to the CNN’s ability to utilize a broader range of information, especially the high spatial resolution of photothermal MIRSI, compared to the RF’s reliance on a limited set of spectral features. This is in line with our previous work which only used 5 wavelengths and resulted in marginally lower accuracy. The enhanced performance with 27 bands, compared to 7, indicates that the reconstruction method not only shortens data collection time but also improves prediction accuracy by offering an expanded spectral data set.…”

“…We demonstrate >95% classification accuracy on a statically robust data set consisting of 98 cervical cancer patients and >40 million data points. Tissue subtype segmentation is a critical step in label-free diagnostics, , and our work lays the foundation for accurate, label-free cervical cancer diagnosis. This work is the first demonstration of applying high-resolution MIRSI to the challenging problem of cervical cancer tissue analysis.…”

“…Both sets included multiple cancer grades. The structures of the RF and CNN classifiers are derived from our published work. − The last TMA row, which contained normal tissue, was used exclusively for testing, while for all other rows, the first five cores were used for training and the remaining cores for testing. This protocol ensures the independence of training and testing data sets, which is a limitation in previously published work …”

Sensitive Discrete Frequency Mid-Infrared Absorption Spectroscopy Using Digitally Referenced Detection

“…As opposed to the continuous, large bandwidth spectrum needed to be acquired by FT- spectrometers, emerging laser technologies offer higher intensities, shorter acquisition times and computer-controlled spectral accuracy. In particular, bright quantum cascade lasers (QCL) have facilitated the development of new-generation rapid imaging, − and promised a better trade-off between analyte localization, scanning speed, and spectral SNR. − QCL-based instruments also allow tuning to specific vibrational bands of interest for fast (time-resolved) sensing − and discrete frequency IR (DFIR) imaging, − while their intrinsically narrow spectral linewidth also provides a unique advantage for applications requiring high spectral resolution. − However, QCL outputs are relatively noisy , – with external cavity (EC), pulsed, and widely tunable configurations being most notably so–yet, they are also the most appropriate design choice for broad-bandwidth mid-IR spectral measurements. − In general, QCL-based spectroscopy is limited by optical power instability that offsets advantages associated with high-intensity throughput. ,, This noise directly affects absorbance measurements and longer acquisitions become necessary to converge on a precise estimate, usually involving hundreds to thousands of pulses. , Alternatively, lock-in demodulation at the pulsing frequency can also be used to estimate average intensity but causes an increase in integration time per spectral band and/or longer spatial dwell time per pixel. Data collected with longer integration times may also suffer from system drift and water vapor variations whereas larger pixel dwell times become the bottleneck for high-resolution, wide-coverage imaging.…”

The promising horizon of deep learning and artificial intelligence in flap monitoring

“…In this era of technologically advanced medicine, the combination of basic clinical examination with Doppler ultrasound remains the gold standard 4 . However, other techniques, such as the application of wireless transmission and cloud computing technologies, ensure that flap perfusion status can be monitored and assessed anytime, anywhere, offering physicians timely intervention and guidance 5 – 7 . Coupled with big data training, there’s potential to further enhance the practicality and precision of various traditional flap monitoring methods (Fig.…”