Determining whether or not a patient has pneumonia based on image classification
-An infection that inflames the air sacs in the lungs, and the fluid accumulation in the lungs can cause a lack of oxygen -The patient struggles to breathe and, without treatment, can succumb to asphyxiation
Bacteria, viruses, fungi, and mold can cause pneumonia
Aspergillus, a common mold that can cause pneumonia:
Mendeley Dataset with 256 pictures of chest x-rays
It contains normal x-rays, x-rays of patients with bacterial pneumonia, and x-rays of patients with viral pneumonia
The viral and bacterial x-ray pictures are split into two classes: normal and pneumonia
I used binary classification to determine if a patient has pneumonia or not.
-I used a convolutional neural network (preferred for image classification) -The diffuse X-ray scattering of the image shows up as fuzziness and the more fuzzy the picture is, that's how the algorithm determines if the patient has pneumonia -Baseline accuracy score of 96-98%
The left image is agar art made by my lab partner Alex. The right image is a photo of Streptococcus viridans on an epithelial cell in the mouth from a gram stain I performed while working in the hospital.
Multi-class algorithm with 3 or 4 nodes, for viral, bacterial, or fungal pneumonia, along with normal lungs (differentiate between all 4 classes) Adapting the technology to help monitor pneumonia, not just diagnosing it
X-rays are not a good tool to monitor the patient’s condition. (It’s too much radiation exposure for the patient.) Consistent lab results and vital signs are how we monitor patients with severe pneumonia Nurses have to count the respiration rate of the patient, but we could set up a camera to do this automatically. The camera could also detect the color of the patient’s skin to see if it turns blue.
Automatic data collection could allow the workers to act more quickly Prevent further exposure of nurses and staff to Covid-19 and other serious diseases, saving more lives overall Patients with level 4 pathogens like Ebola would be well-contained in their room, with very little intervention needed from nurses With the help of data science algorithms, cameras, and robots to physically assist the patients, we could save the lives of more healthcare workers