Ultrasound (US) imaging is used in the diagnosis and monitoring of COVID-19 and breast cancer. The presence of Speckle Noise (SN) is a downside to its usage since it decreases lesion conspicuity. Filters can be used to remove SN, but they involve time-consuming computation and parameter tuning. Several researchers have been developing complex Deep Learning (DL) models (150,000–500,000 parameters) for the removal of simulated added SN, without focusing on the real-world application of removing naturally occurring SN from original US images. Here, a simpler (<30,000 parameters) Convolutional Neural Network Autoencoder (CNN-AE) to remove SN from US images of the breast and lung is proposed. In order to do so, simulated SN was added to such US images, considering four different noise levels ($\sigma$ = 0.05, 0.1, 0.2, 0.5). The original US images (N = 1227, breast + lung) were given as targets, while the noised US images served as the input. The Structural Similarity Index Measure (SSIM) and Peak Signal-to-Noise Ratio (PSNR) were used to compare the output of the CNN-AE and of the Median and Lee filters with the original US images. The CNN-AE outperformed the use of these classic filters for every noise level. To see how well the model removed naturally occurring SN from the original US images and to test its real-world applicability, a CNN model that differentiates malignant from benign breast lesions was developed. Several inputs were used to train the model (original, CNN-AE denoised, filter denoised, and noised US images). The use of the original US images resulted in the highest Matthews Correlation Coefficient (MCC) and accuracy values, while for sensitivity and negative predicted values, the CNN-AE-denoised US images (for higher $\sigma$ values) achieved the best results. Our results demonstrate that the application of a simpler DL model for SN removal results in fewer misclassifications of malignant breast lesions in comparison to the use of original US images and the application of the Median filter. This shows that the use of a less-complex model and the focus on clinical practice applicability are relevant and should be considered in future studies.

The admission of COVID-19 patients to the Intensive Care Unit (ICU) is largely dependent on illness severity, yet no standard criteria exist for this decision. Here, lung ultrasound (LU) data, blood gas analysis (BGA), and clinical parameters from venous blood tests (VBTs) were used, along with machine-learning (ML) models to predict the need for ICU admission. Data from fifty-one COVID-19 patients, including ICU admission status, were collected. The information from LU was gathered through the identification of LU findings (LUFs): B-lines, irregular pleura, subpleural, and lobar consolidations. LU scores (LUSs) were computed by summing predefined weights assigned to each LUF, as reported in previous studies. In addition, individual LUFs were analyzed without calculating a total LUS. Support vector machine models were built, combining the available clinical data to predict ICU admissions. The application of ML models to individual LUFs outperformed standard LUS approaches reported in previous studies. Moreover, combining LU data with results from other medical exams improved the area under the receiver operating characteristic curve (AUC). The model with the best overall performance used variables from all three exams (BGA, LU, VBT), achieving an AUC of 95.5%. Overall, the results demonstrate the significant role of ML models in improving the prediction of ICU admission. Additionally, applying ML specifically to LUFs provided better results compared to traditional approaches that rely on traditional LUSs. The results of this paper are deployed on a web app.

Background/Objectives: Intensity discretization influences radiomic measurements, but its effects in optical coherence tomography (OCT) of choroidal lesions remain insufficiently characterized. We evaluated fixed bin width (FBW) and fixed bin number (FBN) strategies with different parameters to inform discretization choices for en face enhanced-depth imaging OCT (EDI-OCT) of stable choroidal nevi. Methods: Ninety-three original-domain radiomic features were extracted from 64 eyes across 11 discretization settings (FBW 2–64; FBN 8–64). We assessed feature values, eye rankings, observer reproducibility, slab-depth dependence, redundancy, and feature-set overlap. Primary absolute-agreement intraclass correlation coefficients (ICCs) used per-eye feature values averaged across six numeric slabs. Histogram entropy and an approximate co-occurrence sampling-density proxy characterized the discretized intensity representation. Results: Seventy-seven features depended on discretization. Median occupied gray levels decreased with increasing bin width from 54 at FBW 2 to five at FBW 25. Eye rankings were similar between neighboring settings (median Spearman ρ , 0.97–0.99), but less concordant in selected FBW–FBN comparisons. Median observer ICCs ranged from 0.84 to 0.91, without clear evidence of differences between settings. Feature sets overlapped less after redundancy reduction. FBW 4 and FBN 32 provided intermediate gray-level resolution and sampling-density proxies, retaining at least 16 occupied gray levels in approximately 99% and 100% of observations, respectively. Conclusions: Intensity discretization substantially alters radiomic texture values, eye rankings, and feature selection. Observer reproducibility alone is insufficient to guide discretization choice. Intensity representation, gray-level compression, and slab depth should also be considered. FBW 4 and FBN 32 emerged as exploratory candidate settings within the evaluated pipeline. External validation and assessment of clinical utility remain necessary.