Team 67 - yeap can

Detection of pneumonia from chest x-ray

Inspiration

Following the recent spike in the number of COVID-19 cases in Singapore, and how it is becoming the new norm, healthcare services have been struggling to cope with the increased workload as resources get strained. Thus, my team was inspired to help the medical staff by building a model that can help to reduce their workload, which will ease the stress on the resources and also allow for better allocation of it, as we sieve out higher-risk patients that hospitals can then channel more resources to for higher chances of recovery.

What it does

Our solution is to train a machine learning model with pneumonia x-ray datasets so that the model can be potentially used to detect cases of COVID-19 pneumonia.

How we built it

We built an image recognition model using transfer learning using a pre-trained model known as InceptionV3.

Challenges we ran into

As we trained the model for more epochs, the model started to show signs of overfitting to the training dataset, which is not ideal as we hoped for better generalization of the model for other data sets as well. Thus, we hope in the future to tackle this problem, by introducing image augmentation and a better spread of dataset, for better generalization of the model.

Accomplishments that we're proud of

We managed to attain a model accuracy of 93.94% based on our test dataset. We are glad that our model is relatively lightweight and fast due to the choice of the pre-trained model we used as it has fewer parameters as compared to other pre-trained models such as VGG16, while not compromising the accuracy of the model. This makes our model more feasible to be deployed on apps for use in healthcare institutions.

What we learned

We learnt how to apply deep learning and machine learning to solve real-world problems. Moreover, we learnt how to debug our way through problems that we faced during this whole process.

What's next for Team 67 - yeap can

We hope to gain further knowledge in the field of data science and AI, by gaining a deeper understanding of the conceptual workings of these models. We hope to also continue joining more hackathons to learn and grow together as a team :)