Volume 10, Issue 3, March– 2025 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25mar722

AI-Enhanced Lung Size

Matching and Eligibility System
Dr. G. Sangeetha1; Vasan K J2; Subash M3; Venu Krishnan S4
1
Assistant Professor; 2,3,4B.E. - CSE Final Year Students
1;2;3;4
Department of Computer Science and Engineering
SRM Valliammai Engineering College, Kattankulathur, Chengalpattu, Tamil Nadu- 603203

Publication Date: 2025/03/21

Abstract: Lung transplantation is a life-saving procedure for patients with end-stage lung disease, and precise donor-
recipient lung size matching is critical to improving transplant success rates. This paper introduces an automated system
that estimates lung size and assesses transplant suitability using chest X-ray images based on computer vision techniques.
The lung segmentation is achieved through a U-Net model, which successfully separates the lung region from X-ray images.
Key anatomical feature landmarks such as width-at-base, width-at-hilum, R-ACPA, R-AMD, L-ACPA, and L-AMD are
identified with computer vision for precise measurement of lung dimensions. The measured lung dimensions are compared
with donor lung sizes to determine transplant suitability. By reducing reliance on subjective assessments and hand
measurements, the technique increases precision, hastens the process of lung matching, and lessens the involvement of
human mistakes. By automating the procedure of eligibility screening, radiologists and transplant surgeons are provided
with reliable, fact-based data to work with, which ultimately enhances decision-making on lung transplantation. This study
helps to show how deep learning and medical imaging technology can assist in enhancing organ transplantation as well as
medical results.

How to Cite: Dr. G. Sangeetha; Vasan K J; Subash M; Venu Krishnan S (2025). AI-Enhanced Lung Size Matching and Eligibility
System. International Journal of Innovative Science and Research Technology, 10(3), 530-535.
https://doi.org/10.38124/ijisrt/25mar722

I. INTRODUCTION used for pediatric lung transplant where size is crucial [6].
Incorporating deep models such as YOLO into lung
Lung transplantation is a life-saving treatment for segmentation has made auto lung sizing more dependable
individuals with end-stage lung illness, including pulmonary with less workload for human calculations as well as
fibrosis, cystic fibrosis, chronic obstructive pulmonary subsequent human errors [7]. Computer-aided lung sizing
disease (COPD), and pulmonary hypertension. But among the with artificial intelligence from chest radiographs emerged as
most critical challenges in lung transplantation is to deliver a novel method for improving donor-recipient matching
an optimal size match between donor and recipient lungs to through objective and quick measurements [8]. Other studies
prevent graft dysfunction and compromised respiratory have also drawn attention to often overlooked findings on
mechanics [1]. Traditional methods of lung size matching chest radiographs, emphasizing the importance of newer
rely on donor height estimation, chest X-rays, and pulmonary imaging modalities in preventing errors in lung
function tests, which may not always prove to be correct [2]. transplantation [9]. The comparison between pulmonary
Advances in deep learning and medical imaging have function tests and CT volumetry has demonstrated that
significantly improved the accuracy of lung segmentation and imaging-based measurements can provide more accurate
anatomical landmark detection, making it possible to estimates of lung volumes, further confirming their
automatically measure lung size [3]. Computed tomography usefulness in pre-transplant evaluation [10].
volumetry has been used for pre-transplant lung size
matching with increased accuracy than conventional methods This study aims to develop an automated lung transplant
[4]. Variability in chest X-ray-based measurements of lung suitability screening system using deep learning and
height and its impact on transplantation results have also been computer vision techniques. The system will segment lung
examined in research with the emphasis on reliable regions automatically from chest X-rays, identify prominent
computational techniques [5]. anatomical landmarks, and compute lung measurements for
comparison with donor lungs for size matching. By utilizing
Machine learning models have even built models in the AI-based techniques, the system hopes to provide a more
recent past to estimate donor lung volume more accurately precise, objective, and efficient alternative to traditional lung
from chest X-ray measurements, which can specifically be

IJISRT25MAR722 www.ijisrt.com 530

Volume 10, Issue 3, March– 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25mar722
size estimation techniques, ultimately improving Tversky loss function and enhanced attention U-Net model
transplantation success rates and patient outcomes. for segmentation of lesions.

II. LITERATURE REVIEW Lastly, Goceri [25] outlined some of the medical image
data augmentation techniques and discussed their
Lung transplantation remains a life-saving procedure effectiveness for improving the performance of deep learning
for patients with end-stage lung diseases. Donor and recipient models in medical imaging applications. These imaging and
lung size matching is a key element in avoiding post- AI developments provide a strong foundation for accurate
transplant complications. There have been studies looking at estimation of lung size and transplantation eligibility
methods of estimating lung size and how this influences determination. This literature review highlights the
transplant success. significant progress achieved in lung size estimation, medical
imaging, and AI-based analysis for lung transplantation. The
Eberlein et al. [11] investigated the effect of lung size integration of AI-based model segmentation with traditional
mismatch on post-operative complications and resource imaging techniques holds the potential to enhance donor-
utilization after bilateral lung transplantation. Their findings recipient matching in the long run, resulting in improved
highlight the significance of precise size estimation in transplantation outcomes as well as patient survival.
preventing primary graft dysfunction and other post-
operative complications [12]. Li et al. [13] highlighted that III. EXISTING SYSTEM
chest X-ray sizing is reflective of pulmonary diagnosis and
body composition and hence, analogous to the risk of primary Lung transplant surgeries need accurate donor-recipient
graft dysfunction, emphasizing the significance of precise lung size matching to enhance post-transplant survival.
lung size estimation. Ouwens et al. [14] proposed using Conventional size-matching techniques are plagued by
predicted total lung capacity (pTLC) for donor-recipient size inconsistencies caused by human measurements and
matching in lung transplantation. This was done to increase radiologist variability. Our system overcomes this limitation
donor-recipient compatibility and reduce the risk of by using AI and computer vision to automate lung size
complications. Similarly, Riddell et al. [15] presented a estimation from chest radiographs. The AI-based method
simplified method for donor-recipient size matching, provides standardization and accuracy, eliminating the
particularly in interstitial lung disease patients, which was subjectivity and variability inherent in human assessments.
found effective in the clinical setting.
The process works on a two-stage system: the first stage
Aside from transplantation, medical imaging has played involves extracting the lung area from chest radiographs
a critical role in the estimation of lung volume. Pierce et al. using a deep learning-based lung segmentation model, and
[16] devised a technique to estimate lung volumes on the the second stage involves identifying feature points to
basis of chest radiographs through the use of information calculate significant lung measurements. The determined
from shape. Schlesinger et al. [20] compared total lung feature points are utilized to calculate six different height and
capacity estimates on chest radiographs and computed width parameters, which play an important role in evaluating
tomography scans with body plethysmography and lung size. These parameters are then utilized to measure
concluded that imaging-based estimates were highly reliable donor and recipient lung sizes, enhancing the matching
to employ clinically. Recent advances in artificial intelligence procedure. The AI-driven method simplifies the evaluation,
(AI) have better improved lung size estimation and disease saving time for radiologists and transplant surgeons to assess
detection. Topalovic et al. [17] demonstrated that AI lung compatibility.
outperforms pulmonologists when reading pulmonary
function tests, thereby its potential application in clinical To ensure the accuracy of the system, we compared its
decision-making. Chassagnon et al. [18] explored the AI-measured lung size with the measurements reported by
application of AI in the diagnosis of lung cancer, highlighting professional radiologists. Based on a sample of 50 lung
the importance of deep learning models in enhancing transplant patients, our model exhibited a measurement error
diagnostic accuracy. of less than 2.5% (less than 7.0 mm), with high reliability
even in difficult cases, including patient rotation,
Mason et al. [19] described the importance of donor- consolidations, or effusions. The strong interrater and
recipient size matching in lung transplant and pointed out that intrarater agreement also attest to the consistency of the
small mismatches would have an enormous effect on patient model, rendering it a good candidate for clinical
outcomes. Lian et al. [21] presented a structure-aware relation application.Our system improves donor lung evaluation
network for thoracic disease segmentation and detection, through objective, repeatable, and automatic measurements,
which could be extended to lung size estimation and removing the need for human calculation. By integrating AI
transplant suitability evaluation. Medical image segmentation into the workflow of lung sizing, healthcare professionals can
has been addressed by state-of-the-art machine learning streamline donor selection, minimize mismatches, and
algorithms. Lin et al. [22] proposed the focal loss for enhance post-transplant survival. Rapid assessment of lung
enhancing dense object detection from cluttered medical size is also useful for emergencies where immediate donor-
images. Salehi et al. [23] proposed the Tversky loss function recipient matching is imperative.
for image segmentation using 3D convolutional networks,
which performed well on medical imaging. Abraham et al. Although our AI-based system has been encouraging,
[24] also enhanced segmentation performance using a focal further testing on a larger dataset is required to validate its

IJISRT25MAR722 www.ijisrt.com 531

Volume 10, Issue 3, March– 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25mar722
efficacy across heterogeneous patient groups. Future studies processed using computer vision algorithms for the detection
can investigate combining other clinical parameters, of major anatomical features to estimate sizes.
including lung volume estimation and functional analysis, to
improve decision-making in lung transplantation. The present In order to measure lung sizes accurately, the system
study identifies the potential of AI in transforming transplant identifies and marks important feature points, including
medicine by introducing automation, enhancing accuracy, width-at-base, width-at-hilum, Right Anterior Costophrenic
and facilitating large-scale studies on lung size matching and Angle (R-ACPA), Right Anterior Mid-Diaphragm (R-AMD),
transplant outcomes. Left Anterior Costophrenic Angle (L-ACPA), and Left
Anterior Mid-Diaphragm (L-AMD). These points are
IV. PROPOSED SYSTEM extremely important while calculating the height and width
of the lungs, which are very significant donor-recipient
Lung transplantation requires precise matching of matching parameters. The system also ensures consistency in
donor and recipient lung size to ensure successful surgery measurement, eliminating variability due to human
with the least possible post-transplant complications. Our subjectiveness in manual assessment.Once lung sizes are
framework uses chest X-ray images and computer vision obtained, these are compared with donor lung size to
techniques to automatically estimate lung size for transplant determine transplant suitability. Automation does away with
suitability screening. The overall objective is to develop a the physical measurements and reduces errors associated with
uniform and efficient system that eliminates human error and manual calculations. This enhances the accuracy and
maximizes matching precision. Through the utilization of efficiency of lung size-matching, thereby simplifying the
deep learning and AI-segmentation-based methods, the screening process. It also allows for quick assessment in
system presents an accurate lung size assessment solution that emergency transplant cases, improving the workflow for
aids radiologists and transplant surgeons in making informed transplant centers.
decisions.
By the integration of AI, deep learning, and computer
The system starts by performing segmentation of the vision, the system described here revolutionizes lung
lungs using the application of a U-Net model, which transplant assessment, optimizing the compatibility matching
represents a category of deep learning architectures utilized and reducing the workload for medical staff. Not only does it
for segmenting medical images. The U-Net is successful in make better donor selection possible, it also stimulates
demarcating areas of the lung from chest radiographs, and innovation in medical image research. The solution proves to
with high accuracy, extracts lung structures. This is a critical be a demonstration of the potential of AI-based solutions in
process since it forms the foundation for further analysis and the health industry, ultimately enhancing the likelihood of
strips away irrelevant background information, focusing on successful lung transplants and improving patient outcomes.
only the lungs. The images of the segmented lungs are

Fig 1 Architecture Diagram

V. METHODOLOGY the model was trained using images from chest X-rays. The
segmented lung images are then analyzed to identify
The suggested technique automates lung sizing in important feature points that are needed for lung size
transplant screening by combining deep learning and estimation. These feature points are width-at-base, width-at-
computer vision methodically. A U-Net model, a well-known hilum, Right Anterior Costophrenic Angle (R-ACPA), Right
deep learning architecture for image segmentation in the Anterior Mid-Diaphragm (R-AMD), Left Anterior
medical field, is used to separate the lungs in the first phase. Costophrenic Angle (L-ACPA), and Left Anterior Mid-
To segment the lung structures with high precision and Diaphragm (L-AMD).
identify the anatomical items that are being operated upon,

IJISRT25MAR722 www.ijisrt.com 532

Volume 10, Issue 3, March– 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25mar722

Fig 2 Lung Segmentation

After the identification of the feature points, accurate VI. EXPERIMENT

lung dimension measurements are calculated and compared
with donor lung sizes to determine transplant suitability. The To test the performance of our AI-based automated lung
process is automated by computer vision algorithms in the sizing system, we performed experiments on a database of
system, providing consistency and minimizing human error. chest X-ray images for which ground truth lung
The measurements are checked against expert radiologist measurements had been annotated by experienced
ratings for finality to ensure reliability. Through the radiologists. The dataset was split into training, validation,
combination of AI-based segmentation and automated and test sets in a ratio of 80:10:10. The U-Net model was
measurement extraction, the method optimizes lung trained for segmenting lungs, and the feature point detection
transplant screening by making the process quicker and more algorithm was tested independently. Accuracy, precision,
efficient. Enhancements in the future could include the recall, and F1-score performance metrics were computed to
inclusion of other clinical parameters and enhanced real-time evaluate the reliability of the model. The confusion matrix
processing to facilitate hospital integration. was utilized to compare the model's classification
performance so that correctly classified feature points
corresponded to expert labels.

Fig 3 Confusion Matrix

IJISRT25MAR722 www.ijisrt.com 533

Volume 10, Issue 3, March– 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25mar722
For lung segmentation, the U-Net model performed model was overall accurate to 94.5%, with an F1-score of
with an average precision of 90.63%, recall of 99.75%, and 93.8%, proving it to be effective at consistently detecting
F1-score of 94.97%, which signified high segmentation critical anatomical markers. These findings affirm that the
accuracy. The confusion matrix revealed excellent agreement system suggested in this paper can generate accurate and
between predicted and actual lung masks with low false standardized measurements of the lung with low error.
positives and false negatives. In feature point detection, the

Fig 4 Model Evaluation

To further confirm the system, we compared the AI- In addition, the combination of deep learning and
derived lung measurements with two expert radiologists' medical imaging in lung transplant assessment demonstrates
reported measurements for a cohort of 50 lung transplant the wider potential of artificial intelligence in medicine.
recipients. The mean error in lung size estimation was <2.5% Through its simplification of the donor-recipient matching
(<7.0 mm), illustrating the model's high reliability. The procedure, this methodology optimizes clinical workflow and
interrater and intrarater agreement between AI predictions facilitates surgeons and radiologists in the making of highly
and expert measurements was strong, illustrating the system's informed, data-driven decisions. Subsequent studies and
clinical utility. The outcomes show that our method testing with larger datasets will continue to perfect this
efficiently minimizes variability in lung size estimation, system and make it an effective organ transplant suitability
guaranteeing precise donor-recipient matching for lung assessment tool. Finally, this research emphasizes the
transplantation. Future enhancement could be aimed at revolutionary nature of AI-based medical imaging in the field
improving the model for complicated cases, like X-rays with of organ transplantation and its potential for changing the
critical lung abnormalities or low image quality. practice of organ transplant procedures.

VII. CONCLUSION REFERENCES

The new automated lung sizing system offers an [1]. H. Greenspan, S. K. Zhou, C. Davatzikos, B. Van
important innovation in lung transplantation through the use Ginneken, J. S. Duncan, J. L. Prince, D. Rueckert, A.
of deep learning and computer vision for accurate donor- Madabhushi, and R. M. Summers, ‘‘ A review of deep
recipient matching. Through the utilization of a U-Net model learning in medical imaging: Imaging traits,
for lung segmentation and identification of important technology trends, case studies with progress
anatomical landmarks, the system reduces the risk of human highlights, and future promises ’’ Proc. IEEE, vol. 109,
error and increases the accuracy of lung dimension estimates. no. 5, pp. 820–838, May 2021, doi:
This automation not only speeds up the transplant assessment 10.1109/JPROC.2021.3054390.
process but also provides a standardized and objective [2]. D. Datta, G. Kumar, K. Swapna, K. K. D. Ramesh, and
method of lung size matching, ultimately enhancing the S. Rajest, ‘‘A review of medical image segmentation
success rate of transplants. Minimizing dependence on algorithms,’’ EAI Endorsed Trans. Pervasive Health
manual measurements and subjective assessments enables Technol., vol. 7, Jul. 2018, Art. no. 169184, doi:
more consistency and efficiency in transplant decision- 10.4108/eai.12-4-2021.169184.
making. [3]. P. Kennedy, X. He, M. H. Hesamian, and W. Jia,
‘‘Deep learning techniques for medical image

IJISRT25MAR722 www.ijisrt.com 534

Volume 10, Issue 3, March– 2025 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165 https://doi.org/10.38124/ijisrt/25mar722
segmentation: Achievements and challenges,’’ J. [16]. Brown DJ, Holmes M, Pierce RJ, Cumming G,
Digit. Imag., vol. 32, no. 4, pp. 582–596, Aug. 2019, Denison DM. “Estimation of lung volumes from chest
doi: 10.1007/s10278-019 00227-x. radiographs using shape information”. Thorax.
[4]. Farahnak K, Harrison MS, and Shepherd HM, 1979;34(6):726–734. doi: 10.1136/thx.34.6.726.
“Utilizing computed tomography volumetry for size [17]. Das N, Burgel PR, Topalovic M, et al. “Artificial
matching prior to lung transplantation: a case series”. intelligence outperforms pulmonologists in the
J Thorac Dis. 2023;15(4):2233–2239. doi: interpretation of pulmonary function tests”. Eur Respir
10.21037/jtd-22-1203. J. 2019;53(4):1801660. doi:
[5]. Vazquez Guillamet MC, Rjob A, Vazquez Guillamet 10.1183/13993003.01660-2018.
R, et al. Uncertainty analysis of chest X-ray lung [18]. De Margerie-Mellon C, Vakalopoulou M, Chassagnon
height measurements and size matching for lung Get al. “Artificial intelligence in lung cancer: current
transplantation. J Thorac Dis. 2022;14(4):1042–1051. applications and perspectives”. Jpn J Radiol.
doi: 10.21037/jtd-21-1755. 2023;41(3):235–244. doi: 10.1007/s11604-022-
[6]. Szugye NA, Moore RA, Roach DJ, “Improved donor 01359-x.
lung size matching by estimation of lung volumes [19]. Batizy LH, Wu J, Mason DP et al. “Matching donor to
based on chest X-ray measurements”. Pediatr recipient in lung transplantation: how much does size
Transplant. 2023;27(8):e14594. doi: matter?” J Thorac Cardiovasc Surg.
10.1111/petr.14594. 2009;137(5):1234–1240.e1. doi:
[7]. Priego-Torres B, Sanchez-Morillo D, León-Jiménez A 10.1016/j.jtcvs.2008.10.024.
Khalili E. “Automatic lung segmentation in chest X- [20]. White DK, Mallory GB, Schlesinger AE, Hildeboldt
ray images using SAM with prompts from YOLO”. CF, Huddleston CB. “Estimation of total lung capacity
IEEE Access. 2024;12:3454188. doi: from chest radiography and chest CT in children:
10.1109/ACCESS.2024.3454188. comparison with body plethysmography”. AJR Am J
[8]. Araki T, Gefter WB, Ismail MK, et al. “Artificial Roentgenol. 1995;165(1):151–154. Doi:
intelligence-driven automated lung sizing from chest 10.2214/ajr.165.1.7785574.
radiographs. Am J Transplant”. 2025;25(1):198–203. [21]. J.Lian,J.Liu,S.Zhang,K.Gao,X.Liu,D.Zhang,andY.Yu
doi: 10.1016/j.ajt.2024.08.015. ,‘‘Astructure aware relation network for thoracic
[9]. Post BA, Hatabu H, Gefter WB. “Commonly missed diseases detection and segmentation,’’ IEEE Trans.
findings on chest radiographs: causes and Med. Imag., vol. 40, no. 8, pp. 2042–2052, Aug. 2021,
consequences”. Chest. 2023;163(3):650–661. doi: doi: 10.1109/TMI.2021.3070847.
10.1016/j.chest.2022.10.039. [22]. T.-Y. Lin, P. Goyal, R. Girshick, K. He, and P. Dollár,
[10]. Kubo T, Shoji T, Fujinaga T, Chen F, Bando T, Date ‘‘Focal loss for dense object detection,’’ IEEE Trans.
H. “Comparison of pulmonary function test and Pattern Anal. Mach. Intell., vol. 42, no. 2, pp. 318–327,
computed tomography volumetry in living lung Feb. 2020, doi: 10.1109/TPAMI.2018. 2858826.
donors”. J Heart Lung Transplant. 2011;30(5):572– [23]. S. S. M. Salehi, D. Erdogmus, and A. Gholipour,
575. doi: 10.1016/j.healun.2010.11.019. ‘‘Tversky loss function for image segmentation using
[11]. Arnaoutakis GJ, Yarmus L, Eberlein M et al. “The 3D fully convolutional deep networks,’’ in Machine
effect of lung size mismatch on complications and Learning in Medical Imaging. Cham, Switzerland:
resource utilization after bilateral lung Springer, 2017, pp. 379–387, doi: 10.1007/978-3-319-
transplantation”. J Heart Lung Transplant. 67389-9_44.
2012;31(5):492–500. doi: [24]. N. Abraham and N. M. Khan, ‘‘A novel focal Tversky
10.1016/j.healun.2011.12.009. loss function with improved attention U-Net for lesion
[12]. Reed RM, Bolukbas S, Eberlein M, et al. “Lung size segmentation,’’ in Proc. IEEE 16th Int. Symp.
mismatch and primary graft dysfunction after bilateral Biomed. Imag. (ISBI), Venice, Italy, Apr. 2019, pp.
lung transplantation. J Heart Lung Transplant”. 683–687, doi: 10.1109/ISBI.2019.8759329.
2015;34(2):233–240. doi: [25]. E. Goceri, ‘‘Medical image data augmentation:
10.1016/j.healun.2014.09.030. Techniques, comparisons and interpretations,’’ Artif.
[13]. Weinkauf J, Hirji A, Li DV, et al. Chest x-ray sizing Intell. Rev., vol. 56, no. 11, pp. 12561–12605, Nov.
for lung transplants reflects pulmonary diagnosis and 2023, doi: 10.1007/s10462-023-10453-z.
body composition and is associated with primary graft
dysfunction risk. Transplantation. 2021;105(2):382–
389. doi: 10.1097/TP.0000000000003238.
[14]. van der Mark TW, van der Bij W, Geertsma A,
Ouwens JP, de Boer WJ, Ko€ eter GH. “Size matching
in lung transplantation using predicted total lung
capacity”. Eur Respir J. 2002;20(6):1419–1422. doi:
10.1183/09031936.02.00294402.
[15]. Ma J, Dunne B, Riddell P et al. “A simplified strategy
for donor-recipient size-matching in lung transplant
for interstitial lung disease”. J Heart Lung Transplant.
2021;40(11):1422–1430. doi: 10.1016/
.healun.2021.06.013.

IJISRT25MAR722 www.ijisrt.com 535