According to a recent study published based on a hospital in Chandigarh, it was revealed that the AI-based approach demonstrated diagnostic performance just like an experienced radiologist. It was seen that the artificial intelligence was as appropriate as an experienced radiologist. The whole story will be discussed in the following article.
An aggressive malignancy with poor detection and a high mortality rate which is Gallbladder cancer. The whole idea of diagnosing it in the early stages, However, if diagnosed early we have a very high mortality rate. The diagnosis is difficult due to the similar imaging features of gallbladder lesions.
Where did this happen?
PGIMER, Chandigarh (Postgraduate Institute of Medical Education and Research) along with IIT, New Delhi (Indian Institute of Technology) were trying to develop a deep learning model for GBC detection. The tech which was working on this experiment was trying to form a DL model which would detect Gallbladder cancer. The research used abdominal ultrasound, along with which it compared the performance with that of an experienced radiologist.
How did the research take place?
The whole of deep learning is to teach AI or rather computers to go through and process data in a particular just like the way the human brain processes it. The method in AI was being created to process the data and create a detection machine for detecting GBC.
A tertiary care hospital of GBC used ultrasound scans from the time frame of August 2019 to June 2021 to create a data base. As an experiment, the DL model in total had a dataset of 233 patients. It was validated on 59 patients and tested on 273 patients.
What is the result of the research?
The diagnosis was put into an accuracy level. It was judged based on sensitivity, specificity and the area under the receiver operating characteristic curve (AUC). Simultaneously, two experienced radiologists were put on board with the same scans and test. They competed with the DL model.
The score of the DL was 92.3 percent in Sensitivity, 74.4 percent in Specificity and lastly 0.887 in AUC for detecting GBC. When this was compared with the results of the experienced radiologists, it was comparable, according to the study.
The whole study did have some limitations which are always present in any experiment or research. It is mainly based on a certain single centered dataset. In order to achieve broader validation, the dataset needs multi-center studies which will need more time and effort. However, the study has a knowledge cutoff date of 2021. Thus, subsequent development in GBC and DL model diagnosis may not be reflected.