Source-linked AI summary
An Approach for Thyroid Nodule Analysis Using Thermographic Images
J. R. González, É. O. Rodrigues, C. P. Damião, C. A. P. Fontes, A. C. Silva, A. C. Paiva, H. Li, C. Du, A. Conci
TL;DR
The paper addresses how thyroid thermography can support analysis of nodules despite image-motion and diagnostic-processing challenges. It reviews acquisition and registration methods, proposes autonomous ROI identification and registration, and explores feature-based classification in a small pilot study. The authors report evidence that four ROI features may help predict whether patients are sick, while emphasizing the need for larger datasets.
Problem
Dynamic thermographic acquisitions can contain patient-motion errors that propagate into later analyses, while physicians must evaluate substantial diagnostic information quickly.
Method
The paper reviews thyroid thermography and proposes acquisition protocols, autonomous registration, ROI identification, image processing, and k-NN analysis of four ROI features.
Results
The study found evidence that four ROI features may be sufficient for a k-NN classifier to predict whether a patient is sick or not.
Takeaways & Limitations
The work presents a preliminary pilot project supporting further investigation of thermography for thyroid-nodule detection and analysis.
Takeaways & Limitations
The study was constrained by a lack of thermographic thyroid images with abnormalities, requiring analysis of the currently available images and future large-scale evaluation.
Abstract
from arXiv · showhide
Thyroid cancer is said to be the second most common type of cancer in female individuals and the third in males by 2030, according to projections. In general, detecting cancer in its early stages improves the chance of survival of the individual. Thermography is a diagnostic tool that has been increasingly used to detect cancer and abnormalities, including that of thyroid. Various methods to segment and detect hot regions in thermograms and, consequently, to detect suspicious tissues present in these images have been proposed. It is well known that medical diagnosis yields a great deal of information. Thus, physicians have to comprehensively analyse and evaluate this information in a short period of time, which is infeasible in most cases. In this work, we perform a general review of thermography , focusing on the thyroid analysis. We propose protocols for image acquisiton and an autonomous registration for thyroid images. We also perform analyses of the image data, which include feature extraction, image processing, and a possible approach for classification of healthy or unhealthy patients. In summary, this work presents a pilot project for detection of tumors in our university hospital, which is part of an effort to support preventive medical actions in our endocrinology department. Under some future adjustments, this project will be submitted for approval by the ethics and research committee of Hospital Universitário Antonio Pedro at Universidade Federal Fluminense (HUAP-UFF) and to the Brazilian Ministry of Health Ethical committee under the name: Evaluation of the importance of thermography to aid diagnosis of thyroid nodules of patients in HUAP-UFF (in Portuguese: Avaliação da importância da termografia no auxílio à investigação diagnóstica de nódulos tireoidianos em pacientes acompanhados no HUAP-UFF).
1 Introduction
The paper motivates thyroid thermography as a way to analyze temperature patterns and abnormalities, while emphasizing that dynamic acquisitions require registration before downstream analysis. It also outlines the need for consistent acquisition and processing conditions when classification is attempted.
- Motivation: Thermal images capture skin-temperature patterns that are generally symmetric across the sagittal plane, while serial variations can indicate abnormality.Thermography can detect temperature and physiologic changes but cannot pinpoint the problem’s internal location.
- Motivation: Thermography detects temperature modifications associated with diseases including cancer, but internal localization remains limited.The Pennes equation is cited as a way to describe how internal body elements contribute to skin thermal distribution.
- Analysis Requirements: Classification methods require labelled examinations with established diagnoses and consistent acquisition, conditions, and computer processing across patients.The labelled exams provide patterns for classifying patients whose diagnoses are unknown.
- Acquisition Protocols: Dynamic acquisition cools the examined area and records a time series while the body returns toward thermal equilibrium; static acquisition records one image.The paper contrasts dynamic and static protocols as alternative acquisition strategies.
- Registration: Breathing-related movements shift body keypoints during time-series acquisition, so registration is needed early to prevent errors propagating through later analyses.The paper identifies registration as a first computational step for relating corresponding points across thermograms.
- Registration: Registration should use a transformation matched to the observed movement, because unnecessarily complex transformations increase processing time and may produce inadequate images.A simple translation is preferred when it is sufficient instead of complex warping or perspective transformations.
2 Literature Review
The literature review organizes thyroid thermography around acquisition, registration, segmentation, feature extraction, and classification. It covers thermal measurement, protocol design, image alignment, transformation models, and prior thyroid-related findings.
- Thermal Imaging: Infrared thermography converts emitted skin radiation into temperature values and can identify abnormalities through temperature differences and asymmetries.The review links asymmetric temperature distributions with physiological changes and neoangiogenesis.
- Analysis Pipeline: The paper frames thermographic analysis as a pattern-recognition pipeline comprising acquisition, storage, ROI and registration, segmentation, feature extraction, and classification or diagnosis.This sequence organizes the review’s discussion of computer vision and medical applications.
- Acquisition Protocols: Thermal acquisition may be static or dynamic, and images may be captured singly, sequentially, or at long intervals to monitor disease progression.Dynamic protocols observe recovery after induced thermal stress, while accompanied captures are typically separated by 3 or 6 months.
- Acquisition Protocols: Acquisition conditions such as room temperature, stabilization time, examination date, patient age, medications, and hormonal therapy can affect thermographic investigations.Prior work also examined temperature recovery after cold stress and other protocol variables.
- Registration: Image registration maps corresponding points between fixed and moving images and optimizes transformations to maximize their correspondence.The process includes transformation, similarity comparison, rasterization, and interpolation.
- Registration: Registration may use intrinsic image content, attached extrinsic markers, or calibrated nonimage-based coordinate systems, each with different automation and movement assumptions.Nonimage-based registration requires calibrated scanners and assumes no patient movement between acquisitions.
- Transformation Models: The review distinguishes rigid transformations, which preserve pairwise distances, from affine transformations that preserve lines, points, and planes.The proposed work considers rigid transformations in R2 using homogeneous coordinates.
- Registration: Autonomous keypoint methods can reduce manual placement but may produce false correspondences and excessive points requiring heuristics.Examples include SIFT, ASIFT, SURF, and Harris Corner Detector.
3 Proposed Approach
The proposed approach evaluates thyroid thermography using ethically approved hospital images acquired with a controlled dynamic protocol. The study uses a small set of healthy and pathological cases and a physical chin marker to support registration and recognition.
- Data and Acquisition: The study analyzes four patients—two healthy and two with pathological thyroid diagnoses or abnormalities—using thermograms acquired at a university hospital.Images were collected with a FLIR ThermaCam S45 under approval from the Universidade Federal Fluminense ethics committee.
- Data and Acquisition: Images were corrected for room humidity and temperature, and patients waited 10 minutes to stabilize metabolism before acquisition.A blue square placed on the chin served as a physical marker for registration and recognition.
1. Protocol
The protocol uses dynamic thermographic acquisition after thermal stress, producing a sequence of 20 images for examination.
- 1. Protocol: The dynamic protocol stops thermal stress when mean skin temperature reaches 29 °C, then begins sequential image acquisition.Images are captured every 15 s over 5 min.
- 1. Protocol: Patients remain seated during thermal-stress application to minimize possible displacements before acquisition begins.
- 1. Protocol: The acquisition produces a sequence of 20 thermal images from an examination.
2. Recommendations to Patients
Patients are asked to avoid several substances and activities before thermographic examination to standardize preparation.
- 2. Recommendations to Patients: At least 2 h before examination, patients should avoid alcohol, caffeine, physical exercises, nicotine, and creams, oils, or deodorants.
- 2. Recommendations to Patients: The preparation recommendations cover beverages, substances, physical activity, and topical products.
- 2. Recommendations to Patients: Patients should complete these restrictions before undergoing the examination.
3. Room Conditions
The examination requires controlled room conditions, patient positioning, and image registration to reduce movement-related processing problems.
- 3. Room Conditions: Room temperature is maintained between 22 and 25 °C, with no windows, openings, directed airflow, or non-fluorescent bulbs.
- 3. Room Conditions: Patients wait 20 min, remove visible accessories, have central temperature checked, and sit with the head slightly tilted back while looking up.
- 3. Room Conditions: Patient movements can misalign thermographic images, and even slight movements may affect subsequent processing.
1. Full-body Modification
Lateral, vertical, and combined patient movements can be corrected using rigid image transformations.
- 1. Full-body Modification: Patient movements include lateral shifts to the left or right and movements toward the top or bottom.
- 1. Full-body Modification: Lateral and vertical movements may occur in combinations.
- 1. Full-body Modification: These tilts can be corrected with rigid transformation using translation and rotation.
2. Local Modifications
The proposed local-processing pipeline addresses head and neck movements by selecting landmarks, applying a Sobel-derived filter, and autonomously identifying the thyroid ROI. The approach improves alignment, but residual artifacts and whole-image rigid registration still limit accuracy.
- Head tilts and other perspective distortions can be corrected with elastic transformations, which are more computationally expensive than rigid transformations.
- Semi-manual registration uses selected landmarks to align images, but residual errors remain because only translation and rotation are applied to the whole image.
- The autonomous method applies a variable-size Sobel-derived filter to identify structures in thyroid thermograms.
- Using t = −40 on the shadow-oriented x-axis, the selected filter segments the lateral neck while removing most remaining image content.
- A fixed-size rectangle is selected as the ROI candidate containing the highest amount of white pixels, but central marker lines can force the rectangle upward.
- The central residuals can be located with a smaller rectangle and removed before repeating the ROI heuristic.
1. ROI before Registration
Extracting the neck ROI before registration focuses alignment on thyroid-relevant anatomy and avoids shoulder-driven errors. The approach remains fast, but intensity-based comparison requires an appropriate similarity measure.
- Registering only the neck region may improve accuracy by excluding shoulders that can remain aligned while the head and neck move.
- ROI-based registration compares intensity values rather than a few keypoints, making it slightly slower but still only a few milliseconds per image pair.
- A suitable similarity measure is needed because ROI intensity values can vary considerably between images; mean difference may perform poorly, whereas Chamfer distance may be more accurate.
2. ROI after Registration
Keeping the full image during registration preserves information and facilitates landmark matching, but it also introduces irrelevant structures. Thresholding can locate nodules in available sick-patient images and supports feature extraction from segmented hot regions.
- Retaining shoulders and other structures makes common landmarks easier to find and preserves information for later processing, but it introduces registration errors from non-neck anatomy.
- A threshold of 209 converts pixels at or above the value to white and lower values to black, and specialists confirmed nodule localization in the available sick-patient images.
- The thresholded hot region supports three intensity features—mean, standard deviation, and maximum—and a fourth feature measuring vertical symmetry.
- The asymmetry measure averages normalized absolute intensity differences between each pixel and its most similar counterpart across the ROI’s vertical axis.
- The reported feature tables include Euclidean-distance comparisons for patients O1 and O3, with asymmetry values of 0.053, 0.063, and 0.070 shown for one comparison.
4 Conclusion
This small-scale preliminary study evaluates infrared imaging for thyroid-nodule detection and analysis through autonomous ROI identification, feature extraction, and classification. The findings provide evidence that four ROI features may help predict whether a patient is sick, while larger-scale analysis remains future work.
- The study evaluates the feasibility of infrared imaging for thyroid-nodule detection and analysis in terms of time, cost, and effect.
- An autonomous thyroid-image ROI identification method is proposed, alongside comparisons of registration and ROI-extraction strategies.
- Four ROI features were extracted and analysed with k-NN, yielding evidence that they may predict whether a patient is sick.
- The methodologies and evidence require analysis on larger datasets because the repository lacked sufficient abnormal thyroid thermographic images.