Deep learning differentiates between healthy and diabetic mouse ears from optical coherence tomography angiography images

Deep learning automatically assesses 2-µm laser-induced skin damage OCT images

The present study proposed a noninvasive, automated, in vivo assessment method based on optical coherence tomography (OCT) and deep learning techniques to qualitatively and quantitatively analyze the biological effects of 2-µm laser-induced skin damage at different irradiation doses. Different doses of 2-µm laser irradiation established a mouse skin damage model, after which the skin-damaged tissues were imaged non-invasively in vivo using OCT. The acquired images were preprocessed to construct the dataset required for deep learning. The deep learning models used were U-Net, DeepLabV3+, PSP-Net, and HR-Net, and the trained models were used to segment the damage images and further quantify the damage volume of mouse skin under different irradiation doses. The comparison of the qualitative and quantitative results of the four network models showed that HR-Net had the best performance, the highest agreement between the segmentation results and real values, and the smallest error in the quantitative assessment of the damage volume. Based on HR-Net to segment the damage image and quantify the damage volume, the irradiation doses 5.41, 9.55, 13.05, 20.85, 32.71, 52.92, 76.71, and 97.24 J/cm² corresponded to a damage volume of 4.58, 12.56, 16.74, 20.88, 24.52, 30.75, 34.13, and 37.32 mm³. The damage volume increased in a radiation dose-dependent manner.

Classification of Tear Film Lipid Layer En Face Maps Obtained Using Optical Coherence Tomography and Their Correlation With Clinical Parameters

PURPOSE

The purpose of this study was to investigate the correlation between the pattern of optical coherence tomography (OCT) en face maps of the tear film lipid layer (TFLL) and lipid layer thickness (LLT), fluorescein breakup time (FBUT), and Schirmer I test values in healthy subjects.

METHODS

Measurements from four clinical data sets were retrospectively analyzed, and TFLL patterns were classified into 3 categories: homogeneous (HOM), wavy (WAV), or dotted (DOT) appearance. Linear mixed model analyses were performed. Intraclass correlation coefficients and index of qualitative variation were computed to investigate interrater and intrasubject variabilities.

RESULTS

For the LLT, a significant difference between HOM and DOT ( P < 0.001, β HOMvsDOT = -6.42 nm) and WAV and DOT ( P = 0.002, β WAVvsDOT = -4.04 nm) was found. Furthermore, the difference between WAV and DOT regarding FBUT ( P < 0.001, β WAVvsDOT = -3.065 seconds) was significant, while no significant differences between any of the classes with respect to the Schirmer I test values were found. An intraclass correlation coefficient of 89.0% reveals a good interrater reliability, and an index of qualitative variation of 60.0% shows, on average, a considerable variability in TFLL pattern class for repeated measurements over 1 hour.

CONCLUSIONS

A new classification method for OCT en face maps of the TFLL is presented. Significant differences between patterns were found with respect to LLT and FBUT. A dotted pattern on dark background appears to be the most stable type of TFLL. The analysis of OCT en face maps of the TFLL provides complimentary information to conventional imaging methods and might give new insights into the characteristics of the TFLL.

A Decision Support System for Diagnosing Diabetes Using Deep Neural Network

Background and Objective

According to the WHO, diabetes mellitus is a long-term condition marked by high blood sugar levels. The consequences might be far-reaching. According to current increases in mortality, diabetes has risen to number 10 among the leading causes of mortality worldwide. When used to predict diabetes using unbalanced datasets from testing, machine learning (ML) classifiers and established approaches for encoding categorical data have exhibited a broad variety of surprising outcomes. Early studies also made use of an artificial neural network to extract features without obtaining a grasp of the sequence information.

Methods

This study offers a deep learning-based decision support system (DSS), utilizing bidirectional long/short-term memory (BiLSTM), to accurately predict diabetic illness from patient data. In order to predict diabetes, the BiLSTM hybrid model was used after balancing the data set.

Results

Unlike earlier studies, this proposed model's trial findings were promising, with an accuracy of 93.07%, 93% precision, 92% recall, and a 92% F1-score.

Conclusions

Using a BILSTM model for classification outperforms current approaches in the diabetes detection domain.