AcneTyper: An automatic diagnosis method of dermoscopic acne image via self-ensemble and stacking.

BACKGROUND: Acne is a skin lesion type widely existing in adolescents, and poses computational challenges for automatic diagnosis. Computer vision algorithms are utilized to detect and determine different subtypes of acne. Most of the existing acne detection algorithms are based on the facial natural images, which carry noisy factors like illuminations. OBJECTIVE: In order to tackle this issue, this study collected a dataset ACNEDer of dermoscopic acne images with annotations. Deep learning methods have demonstrated powerful capabilities in automatic acne diagnosis, and they usually release the training epoch with the best performance as the delivered model. METHODS: This study proposes a novel self-ensemble and stacking-based framework AcneTyper for diagnosing the acne subtypes. Instead of delivering the best epoch, AcneTyper consolidates the prediction results of all training epochs as the latent features and stacks the best subset of these latent features for distinguishing different acne subtypes. RESULTS: The proposed AcneTyper framework achieves a promising detection performance of acne subtypes and even outperforms a clinical dermatologist with two-year experiences by 6.8% in accuracy. CONCLUSION: The method we proposed is used to determine different subtypes of acne and outperforms inexperienced dermatologists and contributes to reducing the probability of misdiagnosis.

AcneGrader: An ensemble pruning of the deep learning base models to grade acne.

BACKGROUND: Acne is one of the most common skin lesions in adolescents. Some severe or inflammatory acne leads to scars, which may have major impacts on patients' quality of life or even job prospects. Grading acne plays an important role in diagnosis, and the diagnosis is made by counting the number of acne. It is a labor-intensive job and it is easy for dermatologists to make mistakes, so it is very important to develop automatic diagnosis methods. Ensemble learning may improve the prediction results of the base models, but its time complexity is relatively high. The ensemble pruning strategy may solve this computational challenge by removing the redundant base models. MATERIALS AND METHODS: This study proposed a novel ensemble pruning framework of deep learning models to accurately detect and grade acne using images. First, we train multi-base models and prune the redundancy models according to the performance and diversity of the models. Then, we construct the new features of the training data by the base models we select in the previous step. Next, we remove the redundancy models further by a feature selection algorithm. Finally, we integrate all the base models by classifiers. The ensemble pruning algorithm was proposed to prune the deep learning base models. RESULTS: The experimental data showed that the ensemble pruned framework achieved a prediction accuracy of 85.82% on the acne dataset, better than the existing studies. To verify our method's effectiveness, we test our method in a skin cancer dataset and greatly outperform the state-of-the-art methods. CONCLUSION: The method we proposed is used to grade acne. Our method's performance outperforms state-of-the-art methods on two datasets, and it can also remove redundancy models to reduce computational complexity.

DiaMole: Mole Detection and Segmentation Software for Mobile Phone Skin Images.

Results: This study developed mole detection and segmentation software DiaMole using mobile phone images. DiaMole utilized multiple deep learning algorithms for the object detection problem and mole segmentation problem. An object detection algorithm generated a rectangle tightly surrounding a mole in the mobile phone image. Moreover, the segmentation algorithm detected the precise boundary of that mole. Three deep learning algorithms were evaluated for their object detection performance. The popular performance metric mean average precision (mAP) was used to evaluate the algorithms. Among the utilized algorithms, the Faster R-CNN could achieve the best mAP = 0.835, and the integrated algorithm could achieve the mAP = 0.4228. Although the integrated algorithm could not achieve the best mAP, it can avoid the missing of detecting the moles. A popular Unet model was utilized to find the precise mole boundary. Clinical users may annotate the detected moles based on their experiences. Conclusions: DiaMole is user-friendly software for researchers focusing on skin lesions. DiaMole may automatically detect and segment the moles from the mobile phone skin images. The users may also annotate each candidate mole according to their own experiences. The automatically calculated mole image masks and the annotations may be saved for further investigations.

Chemical vapor deposition of thin crystals of layered semiconductor SnS2 for fast photodetection application.

Layered two-dimensional (2D) semiconductors, such as MoS(2) and SnS(2), have been receiving intensive attention due to their technological importance for the next-generation electronic/photonic applications. We report a novel approach to the controlled synthesis of thin crystal arrays of SnS(2) at predefined locations on chip by chemical vapor deposition with seed engineering and have demonstrated their application as fast photodetectors with photocurrent response time ∼ 5 µs. This opens a pathway for the large-scale production of layered 2D semiconductor devices, important for applications in integrated nanoelectronic/photonic systems.

Multiscale simulation on a light-harvesting molecular triad.

We have investigated the effect of solvation and confinement on an artificial photosynthetic material, carotenoid-porphyrin-C(60) molecular triad, by a multiscale approach and an enhanced sampling technique. We have developed a combined approach of quantum chemistry, statistical physics, and all-atomistic molecular dynamics simulation to determine the partial atomic charges of the ground-state triad. To fully explore the free energy landscape of triad, the replica exchange method was applied to enhance the sampling efficiency of the simulations. The confinement effects on the triad were modeled by imposing three sizes of spherocylindrical nanocapsules. The triad is structurally flexible under ambient conditions, and its conformation distribution is manipulated by the choice of water models and confinement. Two types of water models (SPC/E and TIP3P) are used for solvation. When solvated by SPC/E water, whose HOH angle follows an ideal tetrahedron, the structural characteristics of triad is compact in the bulk systems. However, under a certain nanosized confinement that drastically disrupts hydrogen bond networks in solvent, the triad favors an extended configuration. By contrast, the triad solvated by TIP3P water shows a set of U-shaped conformations in the confinement. We have shown that a slight structural difference in the two water models with the same dipole moment can have great distinction in water density, water orientation, and the number of hydrogen bonds in the proximity of a large flexible compound such as the triad. Subsequently, it has direct impact on the position of the triad in a confinement as well as the distribution of conformations at the interface of liquid and solid in a finite-size system.