Ovarian lesions are common and often incidentally detected. A critical shortage of expert ultrasound examiners has raised concerns of unnecessary interventions and delayed cancer diagnoses. Deep learning has shown promising results in the detection of ovarian cancer in ultrasound images; however, external validation is lacking. In this international multicenter retrospective study, we developed and validated transformer-based neural network models using a comprehensive dataset of 17,119 ultrasound images from 3,652 patients across 20 centers in eight countries. Using a leave-one-center-out cross-validation scheme, for each center in turn, we trained a model using data from the remaining centers. The models demonstrated robust performance across centers, ultrasound systems, histological diagnoses and patient age groups, significantly outperforming both expert and non-expert examiners on all evaluated metrics, namely F1 score, sensitivity, specificity, accuracy, Cohen’s kappa, Matthew’s correlation coefficient, diagnostic odds ratio and Youden’s J statistic. Furthermore, in a retrospective triage simulation, artificial intelligence (AI)-driven diagnostic support reduced referrals to experts by 63% while significantly surpassing the diagnostic performance of the current practice. These results show that transformer-based models exhibit strong generalization and above human expert-level diagnostic accuracy, with the potential to alleviate the shortage of expert ultrasound examiners and improve patient outcomes.
23 February 2022 in IEEE Transactions on Biomedical Engineering
The ability to detect dental caries at early stages lies at the heart of minimal intervention dentistry, enabling the curing or arresting of carious lesions before they advance to the cavity stage. Enhanced truncated-correlation photothermal coherence tomography (eTC-PCT) using mid-wave infrared (MWIR) cameras has recently been shown to offer tomographic visualization of early caries. The tomographic slicing ability of such systems, however, is believed to be limited by direct radiative thermal emission through the translucent dental enamel in the 3-5 µm MWIR spectral range. Such radiative emissions can dominate the delayed conductive thermal contributions needed for tomographic reconstruction of internal dental defects. It has been hypothesized that long-wave infrared (LWIR) eTC-PCT systems may offer better tomographic performance by taking advantage of the intrinsic attenuation of direct radiative emission by dental enamel in the LWIR spectral range, enabling more effective delayed conductive thermal contributions from subsurface caries. More than an order of magnitude lower cost of the system is another key attribute of LWIR eTC-PCT which can open the door for downstream translation of the technology to clinics. In this report, we offer a systematic comparison of the performance/effectiveness of caries detection with LWIR and MWIR eTC-PCT systems for detecting natural caries, bacterial caries, and artificially demineralized enamel surfaces. Our results suggest that the low-cost LWIR based eTC-PCT system provides 3D visualization and 2D slice-by-slice images of early caries and internal micro-cracks similar to those obtained from the more expensive MWIR-based eTC-PCT system, albeit with ∼1.3dB lower signal-to-noise ratio.
Tooth decay, or dental caries, is a widespread and costly disease that is reversible when detected early in its formation. Current dental caries diagnostic methods including X-ray imaging and intraoral examination lack the sensitivity and specificity required to routinely detect caries early in its formation. Thermophotonic imaging presents itself as a highly sensitive and non-ionizing solution, making it suitable for the frequent monitoring of caries progression. Here, we utilized a treatment protocol to produce bacteria-induced caries lesions. The lesions were imaged using two related three-dimensional photothermal imaging modalities: truncated correlation photothermal coherence tomography (TC-PCT) and its enhanced modification eTC-PCT. In addition, micro-computed tomography (μ-CT) and visual inspection by a clinical dentist were used to validate and quantify the severities of the lesions. The observational findings demonstrate the high sensitivity and depth profiling capabilities of the thermophotonic modalities, showcasing their potential use as a non-ionizing clinical tool for the early detection of dental caries.
SignificanceIn the analysis of two-layered turbid dental tissues, the outer finite-thickness layer is modeled by an optical transport coefficient distinct from its underlying semi-infinite substrate layer. The optical and thermophysical parameters of healthy and carious teeth across the various wavelengths were measured leading to the determination of the degree of reliability of each of the fitted parameters, with most reliable being thermal diffusivity and conductivity, enamel thickness, and optical transport coefficient of the enamel layer. Quantitative pixel-by-pixel images of the key reliable optical and thermophysical parameters were constructed.AimWe introduced a theoretical model of pulsed photothermal radiometry based on conduction-radiation theory and applied to quantitative photothermal detection and imaging of biomaterials. The theoretical model integrates a combination of inverse Fourier transformation techniques, avoiding the conventional cumbersome analytical Laplace transform method.ApproachTwo dental samples were selected for analysis: the first sample featured controlled, artificially induced early caries on a healthy tooth surface, while the second sample exhibited natural defects along with an internal filling. Using an Nd:YAG laser and specific optical parametric oscillator (OPO) wavelengths (675, 700, 750, and 808 nm), photothermal transient signals were captured from different points on these teeth and analyzed as a function of OPO wavelength. Measurements were also performed with an 808-nm laser diode for comparison with the same OPO wavelength excitation, particularly for the second sample with natural defects.ResultsThe findings demonstrated that the photothermal transient signals exhibit a fast-decaying pattern at shorter wavelengths due to their higher scattering nature, while increased scattering and absorption in the carious regions masked conductive and radiative contributions from the underlayer. These observations were cross-validated using micro-computed tomography, which also enabled the examination of signal patterns at different tooth locations.ConclusionsThe results of our study showed the impact of optical and thermal characteristics of two-layered turbid dental tissues via an inverse Fourier technique, as well as the interactions between these layers, on the patterns observed in depth profiles.