Building AI systems!

The lung is a critical organ for blood gas exchange. Early lung disease detection is often hindered by subtle symptoms. The diagnosis typically requires expert analysis of lung image scans, where both conventional methods and advanced machine learning (ML) and deep learning (DL) techniques are employed for scan segmentation and disease detection. However, it is a challenge to develop reliable computational models, due to the scarcity of high-quality data. With a major emphasis on lung disease classification and segmentation performance, this work presents a novel data quality assessment technique specifically designed for lung image datasets. The proposed pipeline combines ensemble learning, unsupervised learning, and generative autoencoders with attention mechanisms (GAME). By including attentional mechanisms in the generative autoencoders, we improve the tool’s ability to locate and prioritise areas of interest in lung images and extract the right features, which increases the accuracy of our algorithms. The pipeline automates quality checks and reduces the need for high-quality data, while reducing human oversight. Because it’s crucial to validate the robustness of the algorithms in our tool, we tested it using lung scans from the NIH-ChestXray and CheXpert datasets, and obtained IOU scores of 0.88 and 0.86, F1 scores of 0.95 and 0.95, and accuracy scores of 0.96 and 0.95, respectively, which shows they can be used as a classification tool as well as a segmentation tool. Given the challenging conditions in which the early diagnosis of lung disease is made, this is a significant step forward in the development of self-sustaining and accurate disease detection systems. The proposed model is made available to the public and the same is accessible via the github link.

Glaucoma is a debilitating disease, that affects tens of millions of people worldwide. It causes optic nerve damage, that leads to a loss of vision. This makes its detection pertinent at an earlier stage. Several machine learning and deep learning approaches have been proposed in the past to classify the retinal fundus images into healthy and diseased. Traditional artificial neural networks (ANNs), which are used to classify images, have been known to be computationally expensive. On the other hand, spiking neural networks (SNNs) are computationally efficient given the binary spike-driven nature. However, datasets for SNN training are hard to find given the event-based data that SNN requires. Here we propose a novel hybrid ANN-SNN based glaucoma classifier, with multiple SNN-LIF layers in between the ANN layers. This does not require encoding the input pixel data into spikes which may be a bottleneck for several datasets. The proposed model has been trained and tested on the G1020 dataset, and achieves an accuracy score of 60%, recall score of 61%, precision score of 59%, and F1 score of 59.8%.

Soil fertility refers to the ability of soil in a particular area to provide favorable chemical, physical and biological characteristics that help the plant in its growth. It is affected by multiple parameters, from the available concentration of Nitrogen in the soil to the concentration of Organic Carbon in the soil. This paper discusses the implementation of an explainable AI (XAI) model based on a Random Forest classifier. The developed model reliably predicts the relative soil fertility of a given soil using its various physiochemical properties, and explain the reasons behind the model’s soil fertility indicator prediction using user friendly graphs. The model shows 97.02% accuracy in comparison with state-of-the-art machine learning models. The paper also discusses applications of developed model in providing possible solutions to further improve upon soil fertility in the short term and long term.