Smart Diagnostics for Pulmonary Healthcare
Researchers at the Dept. of Electronics and Electrical Communications Engineering at IIT Kharagpur have developed a decision support system to diagnose malignant and other diseased tissues in the lungs. While one system can refer to CT scan images to detect lung nodules and test them for the possibility of malignancy, a second software can detect Interstitial Lung Disease (ILD) patterns in chest HRCT images.
“Biopsy especially in the lungs is a critical process, hence conducted only after initial medical analysis is done by expert radiologists. The developed systems use noninvasive and comparatively affordable methods of image analysis that would aid the radiologists to identify malignancy by reading growth in the lung nodules. The other system will help identify interstitial disease patterns in HRCT images depicting the lung tissue texture,” explained lead researcher Prof. Sudipta Mukhopadhyay.
“The novelty of the system lies in its India-centric reference point i.e. the medical image scan database used for reference is sought from the Indian patient population. We worked with Prof. Khandelwal and his team from PGIMER Chandigarh for data ground truth and clinical data. Also, foreign database such as LIDC-IDRI and MedGIFT ILD database has been used. The biopsy cases were primarily taken from PGIMER,” explained researcher Shrikant Mehre.
The malignancy detecting tool detects the lung nodule, segment the nodule, and provides a way to modify segmentation, retrieve similar nodules from the database with their report and assess the chance of malignancy of the query nodule based on the retrieval results. The ILD tool is developed by incorporating feedback from expert radiologists to make it easy to use for non-tech savvy clinicians. The software is equipped with necessary modules such as automatic segmentation of lung boundary and pathological region within lung area, provision to modify the boundary, retrieving similar segments from the database with their report and assess the probability of the pathological segment to be a particular ILD category based on the retrieval results. The mapping of disease is performed by doctors based on the ILD pattern and clinical inputs.
“We have successfully tested both software systems at AIIMS Delhi. Prof. Ashu Seith Bhalla and her team provided the neutral test site required for the validation. Currently, lung nodule detection rate and classification rate is 86% and 87%, respectively, and the success rate for ILD classification is 84%. We are working towards further improvements in order to conduct clinical trials on bigger sample sizes,” said another researcher Mandar Kale.
With the growing cases of cancer and other respiratory diseases in India, the need for skilled radiologists is expected to grow exponentially in near future. Budding radiologists will be highly benefited in learning from previous images stored in Picture Archival and Communication System (PACS) and reports in Radiological Information System (RIS) on their own and to help practicing radiologists in differential diagnosis using the CBIR based Computer Aided Diagnosis (CAD) system.
The research has been reported in more than 13 international journals and 19 international conferences through its various stages of progress.