Thermal interpretation procedure for the adjunct detection of thyroid pathologies

González, Francisco; González, Raymundo; Rios, Julian; Rodríguez, Eduardo Hopkins; Milián, Ximena

doi:10.1117/12.2576786

Cited by 1 publication

(1 citation statement)

References 0 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The dataset consists of 640×480 one-channel images, along with their respective masks providing information about the location of the thyroid nodule. It is important to notice that thermogram images were processed subtracting its mirrored image, following the analysis of [14]. The dataset was split into 80% for training and 20% for testing, with cross-validation used to evaluate the network's performance.…”

Section: Methods 21 Databasementioning

confidence: 99%

Thermal detection of thyroid pathologies using artificial neural networks

Flores Nuñez,

Jurado-Aguirre,

Pérez-Nájera

et al. 2023

Infrared Remote Sensing and Instrumentation XXXI

View full text Add to dashboard Cite

Thyroid cancer is the most common type of neoplasm in the head and neck regions, thermal imaging has been used successfully for medical diagnostic purposes in areas such as breast cancer, wound care, vascular diseases, skin cancer and eye diseases. The thyroid is a richly vascularized gland that is located close to the skin, its hyper or hypo activity modifies the temperature pattern of the neck making thermography a good candidate to evaluate possible pathologies by digital infrared thermography. In this work an Artificial Neural Network based on the Nested U-Net architecture is trained and used to detect thyroid pathologies. Two thermographic databases were used to train the artificial neural network, the dataset was split into 80% for training and 20% for testing, with cross-validation used to evaluate the network's performance. The error between the predicted masks and the actual masks is calculated with the combination of Binary Cross Entropy (BCE) Loss and Dice Loss using Adam algorithm as training rule, with a cosine annealing schedule during 1500 epochs with a learning rate of 0.003 and a batch size of 6. Results show that when calculating the correlation coefficient between these deltas and the results of thyroid ultrasound expressed in the ACR TI-RADS classification, a high degree of correlation is obtained.

show abstract

Section: Methods 21 Databasementioning

confidence: 99%