"Evaluation of AI performance in detecting lung opacities in chest radiographs and charecterizing them into diffrent subgroups (atelectasis, calcification, cardiomegaly, fibrosis, mediastinal widening, nodule and pleural effusion)."

Chest radiography, or X-ray, one of the most common imaging examsworldwide, is performed to help diagnose the source of symptoms like cough,fever and pain.

Radiologicevaluation of lung lesions can play many roles including primary detection,narrowing of the differential diagnosis, treatment or surgical planning, andpost treatment surveillance.

The chest radiographic findings alone are nonspecific andnot sufficient for the definitive diagnosis of pulmonary infection, but it playan important role in patients who are critically ill in intensive care unit, incombination with clinical findings radiographs can substantially improve theaccuracy of diagnosis in this disease. Serial radiographs repeated bedside inthe critical care setting can also help monitor progression / improvement andthus help with clinical management of the patients**(Patino Gonzalez et al., 2020)**.

Chest radiographs demonstrate normal findings and alsohelp characterizing them into different subgroups (atelectasis, calcification,cardiomegaly, fibrosis, mediastinal widening, nodule and pleural effusion).) Itcan also aid in characterizing the disease into benign and malignant.

Radiograph not only helps in detecting lung opacities butalso aids in detecting multiple medical and surgical emergencies such asspontaneous pneumothorax,pneumothorax can precipitate a life-threateningemergency due to lung collapse and respiratory or circulatory distress.Pneumothorax is typically detected on chest X-ray (O’Connor & Morgan, 2005).

Large data being processed and due to heavy workload intertiary centers, there is high chance of missing findings in reporting andthere can be delay in reporting.

There is need of system which will assess the radiographcorrectly and with less time, which not only raise the momentum in emergencycases but also helps in better utilization of manpower.

 Applicationof AI in imaging is evolving and could be valuable tool in improving workflowand workforce efficiency. Many AI algorithms are available in the radiologydepartments and critical care suites which are designed to identify and flagthe findings to the radiologist for second review**(Yasaka & Abe, 2018)**.

Thenew era of artificial intelligence (AI) has introduced revolutionarydata-driven analysis paradigms that have led to significant advancements ininformation processing techniques in the context of clinical decision-supportsystems. These advances have created significant impact on rapid diagnosis andhas raised the momentum in computational medical imaging applications and alsoaids in new precision medicine research areas**(Trivizakis et al., 2020)**.

The AIM of current study is

(a)Imageinterpretation by “AI alone†has similar accuracy to image interpretation by “radiologist alone.â€

(b) image interpretation by “ AI+ radiologist†increase the accuracy and reduces the interpretation time compared to image interpretation by “radiologist aloneâ€.

(c) The confidence in diagnostic is increased for radiologists when AI is used in clinical practice.

The study will be conducted in 3 phases:

Phase one:

The performance of AI alone will be evaluated on is set of prior 1000(A+B+C) radiographs composed of:

• SubsetA 300 chest radiograph randomly selected from a patient population aged 15 plus with respiratory symptoms.
• Subset B 300 chest radiograph randomly selected from a patient population age of 15 plus diagnosed with lung cancer.
• Subset C 400 chest radiograph randomly selected from a patient population age and 15 plus with active tuberculosis symptoms.

The performance of TCS will be evaluated as follows:

• For detection of lung capacity on subject A
• For detection of nodules on subject B
• For detection of atelectasis, calcification, cardiomegaly, fibrosis, mediastinal widening, nodule, and pleural effusion in subjects A, B, and C.

PHASE 2:

The accuracy and the interpretation time of “radiologist†and “AI+ radiologist†for detecting atelectasis, calcification, cardiomegaly, fibrosis, mediastinal widening, nodule and pleural effusion. The set of1000 chest radiographs will be randomly selected from the CXR exams performed in the outpatient department and in the emergency department at the institution during day 4 months after the installation of AI tools.

PHASE 3:

After phase 2 consultants and residents will fill a survey to indicate their relative confidence in picking findings with and without AI and perceived advantages/disadvantages of reading with AI.