Letter to ‘Automated CT image prescription of the gallbladder using deep learning: Development, evaluation, and health promotion’

Abstract

We would like to comment on “Automated CT image prescription of the gallbladder using deep learning: Development, evaluation, and health promotion.¹” This study created an automatic detection system for acute cholecystitis (AC) that can recognize gallbladders from CT scans of patients and controls. It detected gallbladders using the VGG-16 architecture and processed them using techniques such as the Flood fill algorithm and centroid calculation, as well as U-Net for picture segmentation and feature extraction. The combination of results from many models aided in the development of an automatic and accurate AC detection system.

However, using the accuracy value to evaluate the system's performance may not be sufficient to reflect the ability to distinguish between cases with different conditions, especially when there is an imbalance of data, such as the difference between AC patients and non-patient controls, or when the data is not evenly distributed, which may cause the accuracy value to not reflect the model's effectiveness in handling more difficult cases. Other indices, such as sensitivity, specificity, and AUC (Area Under Curve) values, can help increase the accuracy of model performance evaluation. The example of this kind of study is the previous publication by Ma et al.²

Furthermore, it should be considered to develop techniques that improve processing in cases with low-quality or noisy images, which may cause the model to misclassify or skip over complex cases. This includes the use of diverse data from various sources, such as adding images from patients with complications or changes in gallbladder characteristics.

Approaches that can learn from various data and adapt to the diversity of CT scans, as well as deep learning approaches, should be applied in future development. Further investigation into the model's capacity to process under multiple scenarios, such as changing operational conditions or patient diversity, will improve the system's robustness and accuracy in practice. This study describes the development of an AI-based AC detection system that can work quickly and accurately; however, further developments in low-quality image processing and the use of more diverse statistical techniques are required to enable this technology to detect the disease more accurately and efficiently in clinical practice.

The authors declare no conflicts of interest.

Approval of the research protocol: Not applicable, there is no involvement of humans or animals.

Informed consent: Not applicable, there is no human subject.

Registry and the registration no. of the study/trial: NA.

Animal studies: NA.