Development of thyroid uptake calculation software using deep learning algorithm

Nuclear Medicine imaging modality is one of the tools to provide functional information of specific organ of the patients. Qualitative and quantitative information for thyroid functional assessment is carried out using this tool for thyroid management. Imaging data can be used by Machine Learning algorithms like Deep Learning (DL), which are evolving rapidly with a great success rate and in a very broad spectrum of applications. These DL models use Neural Engines and Neural Networks to identify the data representations based on mathematics and algorithms which is called threshold logic. The objective of this study is to apply a practical frame-work of the automated detection of Region of Interest (ROI) in the image for thyroid uptake calculation. Total 36 patients, referred for the thyroid uptake and scan study, were included in this study. The patients were administered 0.925 MBq (25 μCi) ¹³¹ I Capsule orally for thyroid scan. The data was transferred in Digital Imaging and Communications in Medicine (DICOM) format from Gamma Camera SPECT system to window based PC. Uptake was calculated using Thyroid Uptake probe and gamma camera at 2 hrs and 24 hrs. However, 24 hrs. uptake data was used for analysis in this study. An artificial intelligence (AI) algorithm is used for automated ROI detection and "Doughnut Background subtraction" on processed DICOM images obtained from the Gamma camera fitted with medium energy all purpose (MEAP) collimator. The uptake measurement obtained by these two methods were compared. The methodology involved in the ROI detection were alpha and gamma channel adjustments and then using canny edge detection which is further passed to an ML (Machine Learning) model to adjust the shape of the ROI according to the specific organ to remove possible radiations coming from nearby organs. The ML model is continuously trained over the dataset from each patient and continues to improve its accuracy. Comparison was done between uptake values obtained with probe system and with the values obtained using DL algorithm on processed DICOM images of the patient. It is found that the correlation coefficient values were lying in the range of 0.946-0.996. Accurate ROIs were drawn successfully with the help of developed algorithm. The developed algorithm for calculation of thyroid uptake is very accurate and reproducible. This is fully automated technique in which inter-operator and the inter-facility variability of ROI setting is completely eliminated. This is a robust technique which is user friendly and will be very useful for uptake calculation, which provides quantitative information of thyroid organ. (author)