Region-aware CT-to-PET image synthesis using deep learning models for accurate lung cancer localization

Accurate functional imaging remains a major challenge in medical diagnostics, particularly in regions with limited access to positron emission tomography (PET). We propose a modular, tumor-aware pipeline that learns PET-associated representations from computed tomography (CT) images using paired CT–PET data as supervisory information. The computer-aided diagnostic (CAD) system consists of four stages: (i) field-of-view alignment using YOLOv11, (ii) synthetic 18 F-FDG PET generation via a ResNet-9 based generator with a patchGAN discriminator (CPGAN), (iii) tumor segmentation with SegResNet to guide region-specific synthesis, and (iv) tumor-aware PET enhancement through tumor superimposition and PET-CT fusion. Experimental evaluations indicate strong performance across multiple tasks. Thoracic region detection achieved mAP50 scores of 99.50% (CT) and 98.50% (PET), while lung segmentation reached 99.20%. 18 F-FDG PET synthesis with CPGAN yielded PSNR 20.94 dB, SSIM 72.94%, and MAE 0.036. Incorporating tumor superimposition substantially improved 18 F-FDG PET quality within tumor regions (SSIM 83.47%, PSNR 25.99 dB, MAE 0.017). After peripheral padding, full-image metrics further improved (SSIM 96.35%, PSNR 33.66 dB, MAE 0.0036). For the segmentation, SegResNet model achieved the highest performance (Dice: 97.06%, IoU: 93.65%). The proposed framework provides an anatomically grounded visualization that can highlight CT regions associated with metabolically active tumors and may assist radiologists in identifying areas warranting further assessment.

Authors

Institutions

Publication Details

Journal
Scientific Reports
Published
2026-11-21
DOI
https://doi.org/10.1038/s41598-026-80981-5
Primary Topic
Medical Imaging Techniques and Applications
Type
article
Field-Weighted Citation Impact
0.00

Funders

Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Region-aware CT-to-PET image synthesis using deep learning models for accurate lung cancer localization

Muhammad Abdul Kadir, Md. Abu Sayed, Md. Sherajul Islam, Muhammad E. H. Chowdhury et al.
1 links
Scientific Reports
Medical Imaging Techniques and Applications
article

Region-aware CT-to-PET image synthesis using deep learning models for accurate lung cancer localization

Muhammad Abdul Kadir, Md. Abu Sayed, Md. Sherajul Islam, Muhammad E. H. Chowdhury, Jeongwon Park, Ashiqur Rahman, G. M. Mahmudur Rahman
article en

Abstract

Accurate functional imaging remains a major challenge in medical diagnostics, particularly in regions with limited access to positron emission tomography (PET). We propose a modular, tumor-aware pipeline that learns PET-associated representations from computed tomography (CT) images using paired CT–PET data as supervisory information. The computer-aided diagnostic (CAD) system consists of four stages: (i) field-of-view alignment using YOLOv11, (ii) synthetic 18 F-FDG PET generation via a ResNet-9 based generator with a patchGAN discriminator (CPGAN), (iii) tumor segmentation with SegResNet to guide region-specific synthesis, and (iv) tumor-aware PET enhancement through tumor superimposition and PET-CT fusion. Experimental evaluations indicate strong performance across multiple tasks. Thoracic region detection achieved mAP50 scores of 99.50% (CT) and 98.50% (PET), while lung segmentation reached 99.20%. 18 F-FDG PET synthesis with CPGAN yielded PSNR 20.94 dB, SSIM 72.94%, and MAE 0.036. Incorporating tumor superimposition substantially improved 18 F-FDG PET quality within tumor regions (SSIM 83.47%, PSNR 25.99 dB, MAE 0.017). After peripheral padding, full-image metrics further improved (SSIM 96.35%, PSNR 33.66 dB, MAE 0.0036). For the segmentation, SegResNet model achieved the highest performance (Dice: 97.06%, IoU: 93.65%). The proposed framework provides an anatomically grounded visualization that can highlight CT regions associated with metabolically active tumors and may assist radiologists in identifying areas warranting further assessment.

Scientific Reports
Khulna University of Engineering and Technology (BD), University of Nevada, Reno (US), University of Ottawa (CA), University of Dhaka (BD), Qatar University (QA)
National Research Foundation of Korea, Institute of Nuclear Medicine and Allied Sciences, Defence Research and Development Organisation
Good health and well-being
Medical Imaging Techniques and Applications
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon