Artificial Intelligence Engineer and Developer in Houston, United States

Our method employs scaled networks for detecting breast cancer in whole slide images. It classifies entire slide images on a patch level into normal, benign, in situ, and invasive tumors.

Our approach yielded an f1-score of 88.9 (±1.7)%, which outperformed the benchmark f1-score of 81.2 (±1.3)% on patch level and achieved an average dice coefficient of 69.8% on 10 whole slide images compared to the benchmark average dice coefficient of 61.5% on BACH dataset.

Similarly, on the Dryad test dataset comprising 173 whole slide images, we achieved an average dice coefficient of 82.7% compared to the previous state-of-art of 76% without fine-tuning on this dataset. We further proposed a method to generate patch-level annotations for the image-level TCGA breast cancer database that will be useful for future deep learning methods.

Despite great success, deep learning models still face a critical obstacle in classifying highly imbalanced real-life data.

Detecting diabetic foot ulcers is fundamental for healthcare specialists to prevent amputations. In this work, we performed multiple experiments to benchmark results on the grand. To adjust the bias of the convolutional neural networks, we also proposed a custom-designed activation layer based on softmax to handle the probability skew of the classes.

We achieved the second position in the validation set with a macro F1 score of 0.593 and the third position in the test set with a macro F1 score of 0.596 for the Diabetic Foot Ulcer Detection 2021 Grand Challenge.

COVID-19 and pneumonia have impacted human life significantly. The number of infected people and deaths is increasing daily due to COVID-19. Rapid COVID-19 detection is vital to control and stop the spread of the disease.

Considering AI can play a significant role in accurately detecting such diseases, EE-RDS conducted a multi-class classification challenge by providing chest X-rays of pneumonia, COVID-19, and regular patients. We proposed PRNet, a novel deep learning pipeline, and achieved 96.3% accuracy, winning the second position on the test set leader board.

Deep learning algorithms have made many breakthroughs and various real-life applications. Computational resources become a bottleneck as the data and complexity of the deep learning pipeline increase.

In this paper, we proposed optimized deep learning pipelines in multiple aspects of training, including time and memory. OpTorch is a machine learning library designed to overcome weaknesses in existing implementations of neural network training. It provides features to train complex neural networks with limited computational resources.

OpTorch achieved the same accuracy as existing libraries on CIFAR-10 and CIFAR-100 datasets while reducing memory usage to approximately 50%. We also explored the effect of weights on total memory usage in deep learning pipelines.

In our experiments, parallel encoding-decoding along with sequential checkpoints result in a much-improved memory and time usage while keeping the accuracy similar to existing pipelines.