Accuracy improvement of quantification information using super-resolution with convolutional neural network for microscopy images

Cell quantification in digital contrast microscopy images with convolutional neural networks algorithm

High Content Screening (HCS) combines high throughput techniques with the ability to generate cellular images of biological systems. The objective of this work is to evaluate the performance of predictive models using CNN to identify the number of cells present in digital contrast microscopy images obtained by HCS. One way to evaluate the algorithm was through the Mean Squared Error metric. The MSE was 4,335.99 in the A549 cell line, 25,295.23 in the Huh7 and 36,897.03 in the 3T3. After obtaining these values, different parameters of the models were changed to verify how they behave. By reducing the number of images, the MSE increased considerably, with the A549 cell line changing to 49,973.52, Huh7 to 79,473.88 and 3T3 to 52,977.05. Correlation analyzes were performed for the different models. In lineage A549, the best model showed a positive correlation with R = 0.953. In Huh7, the best correlation of the model was R = 0.821, it was also a positive correlation. In 3T3, the models showed no correlation, with the best model having R = 0.100. The models performed well in quantifying the number of cells, and the number and quality of the images interfered with this predictive ability.

Double paths network with residual information distillation for improving lung CT image super resolution

Objective

Medical image analysis is particularly important for doctors to differential diagnosis of diseases. Due to the outbreak of COVID-19, how to diagnose COVID-19 accurately has become a key issue. High-resolution lung CT images can provide more diagnostic information, so there is an urgent need to develop a super-resolution method to improve the resolution of medical images.

Methods

In this paper, a method based on double paths with residual information distillation for medical images super resolution (DRIDSR) is established. In the low-frequency path, shallow convolutional network is used to get low-frequency features, while in the high-frequency path, a residual information distillation module (RIDM) is designed to obtain clearer high-frequency features. RIDM cascades multiple residual blocks, and uses the output of each residual block as the input of IDB for further information distillation. Finally, it merges the information left by multiple IDBs as output.

Results

The proposed method is tested on the public dataset COVID-CT. The DRIDSR reconstruction quality of the algorithm is higher than that of the SRCNN, ESPCN, VDSR, IMDN and PAN method (+2.21 dB, +2.41 dB, +1.42 dB, +0.43 dB, +0.54 dB improvement, respectively) at × 3 upscale factor and (+2.35 dB, +2.17 dB, +1.59 dB, +0.48 dB, +0.56 dB increase, respectively) at ×4 upscale factor. While the number of parameters and analysis time of our model are reduced.

Conclusions

It is demonstrated that DRIDSR network can obtain better performance and better HR medical images than several state-of-the-art SR methods in terms of objective indicators and subjective evaluation.

AptaNet as a deep learning approach for aptamer-protein interaction prediction

Aptamers are short oligonucleotides (DNA/RNA) or peptide molecules that can selectively bind to their specific targets with high specificity and affinity. As a powerful new class of amino acid ligands, aptamers have high potentials in biosensing, therapeutic, and diagnostic fields. Here, we present AptaNet-a new deep neural network-to predict the aptamer-protein interaction pairs by integrating features derived from both aptamers and the target proteins. Aptamers were encoded by using two different strategies, including k-mer and reverse complement k-mer frequency. Amino acid composition (AAC) and pseudo amino acid composition (PseAAC) were applied to represent target information using 24 physicochemical and conformational properties of the proteins. To handle the imbalance problem in the data, we applied a neighborhood cleaning algorithm. The predictor was constructed based on a deep neural network, and optimal features were selected using the random forest algorithm. As a result, 99.79% accuracy was achieved for the training dataset, and 91.38% accuracy was obtained for the testing dataset. AptaNet achieved high performance on our constructed aptamer-protein benchmark dataset. The results indicate that AptaNet can help identify novel aptamer-protein interacting pairs and build more-efficient insights into the relationship between aptamers and proteins. Our benchmark dataset and the source codes for AptaNet are available in: https://github.com/nedaemami/AptaNet .