Document Type
Event
Start Date
23-4-2023 5:00 PM
Description
Abstract—Bacterial species identification is an essential step in diagnosing diseases caused by bacterial attacks. Effective prescription to cure these diseases depends on accurate bacterial species identification. Faster identification along with accuracy is essential because some bacteria grow fast in the human body. However, bacterial species classification using a laboratory environment through traditional approaches is time-consuming, and it depends on human expertise, which is not immune to human error. Well-trained and experienced microbiologists demonstrate the probability of preparing bacterial species identification reports with lower error rates. However, hiring experienced microbiologists is expensive. Convolutional Neural Network (CNN)-based automatic bacterial species classification system is a potential innovative alternative to traditional laboratory methods, which is faster and cheaper. However, there are multiple challenges associated with developing a CNN-based bacterial species classifier. This paper proposes an innovative CNN-based framework, BactiFind, to overcome the obstacles and automatically identify 33 different bacterial species and generate reports for physicians. It classifies bacterial species with 96.42% accuracy, 97.13% precision, 97.25% recall, and 3.58% error rate. The well-optimized network architecture, innovative bypass layer, and effective image augmentation method have facilitated this outstanding performance which has been further validated using state-of-the-art CNN evaluation methods. Index Terms—Bacterial species classification, Convolutional Neural Network, image augmentation, network optimization, bacterial image transformation.
UR-390 BactiFind: A Novel CNN-based Framework to Classify Bacterial Species
Abstract—Bacterial species identification is an essential step in diagnosing diseases caused by bacterial attacks. Effective prescription to cure these diseases depends on accurate bacterial species identification. Faster identification along with accuracy is essential because some bacteria grow fast in the human body. However, bacterial species classification using a laboratory environment through traditional approaches is time-consuming, and it depends on human expertise, which is not immune to human error. Well-trained and experienced microbiologists demonstrate the probability of preparing bacterial species identification reports with lower error rates. However, hiring experienced microbiologists is expensive. Convolutional Neural Network (CNN)-based automatic bacterial species classification system is a potential innovative alternative to traditional laboratory methods, which is faster and cheaper. However, there are multiple challenges associated with developing a CNN-based bacterial species classifier. This paper proposes an innovative CNN-based framework, BactiFind, to overcome the obstacles and automatically identify 33 different bacterial species and generate reports for physicians. It classifies bacterial species with 96.42% accuracy, 97.13% precision, 97.25% recall, and 3.58% error rate. The well-optimized network architecture, innovative bypass layer, and effective image augmentation method have facilitated this outstanding performance which has been further validated using state-of-the-art CNN evaluation methods. Index Terms—Bacterial species classification, Convolutional Neural Network, image augmentation, network optimization, bacterial image transformation.