Longitudinal Assessment of Lung Lesion Burden in CT (2504.06924v1)

Abstract: In the U.S., lung cancer is the second major cause of death. Early detection of suspicious lung nodules is crucial for patient treatment planning, management, and improving outcomes. Many approaches for lung nodule segmentation and volumetric analysis have been proposed, but few have looked at longitudinal changes in total lung tumor burden. In this work, we trained two 3D models (nnUNet) with and without anatomical priors to automatically segment lung lesions and quantified total lesion burden for each patient. The 3D model without priors significantly outperformed ($p < .001$) the model trained with anatomy priors. For detecting clinically significant lesions $>$ 1cm, a precision of 71.3\%, sensitivity of 68.4\%, and F1-score of 69.8\% was achieved. For segmentation, a Dice score of 77.1 $\pm$ 20.3 and Hausdorff distance error of 11.7 $\pm$ 24.1 mm was obtained. The median lesion burden was 6.4 cc (IQR: 2.1, 18.1) and the median volume difference between manual and automated measurements was 0.02 cc (IQR: -2.8, 1.2). Agreements were also evaluated with linear regression and Bland-Altman plots. The proposed approach can produce a personalized evaluation of the total tumor burden for a patient and facilitate interval change tracking over time.

An Expert Review of "Longitudinal Assessment of Lung Lesion Burden in CT"

The paper "Longitudinal Assessment of Lung Lesion Burden in CT" authored by Tejas Sudharshan Mathai, Benjamin Hou, and Ronald M. Summers addresses a critical issue in radiology related to lung cancer diagnosis and monitoring. Lung cancer remains a significant cause of mortality in the United States, and early detection through imaging techniques like CT is vital for improving patient outcomes. The paper focuses on advancing the segmentation and longitudinal assessment of lung lesion burden using deep learning methodologies.

In this research, two 3D models based on nnUNet architectures were trained to segment lung lesions: one utilizing anatomical priors, derived from the TotalSegmentator tool, and one without such priors. The paper evaluates the automatic segmentation of lung lesions from longitudinal CT studies to monitor changes in tumor burden over time.

Key Results and Analysis

The authors report that the model without anatomical priors significantly outperforms the model with priors in various metrics, including a precision of 71.3% and an F1-score of 69.8% for detecting lesions greater than 1 cm. In terms of segmentation, the model without priors achieved a Dice coefficient of 77.1 and a Hausdorff distance error of 11.7 mm. These results suggest that the absence of anatomical priors may lead to improved detection and segmentation performance, although this model also produced false positives outside the lung region.

The practical implications of this work are noteworthy. By providing automated, accurate quantifications of lung lesion burdens, this method has the potential to streamline radiological workflows, reduce subjectivity in lesion measurement, and provide critical data for the treatment planning of patients diagnosed with lung cancer. Such automated systems could greatly enhance the consistency of measurements, which is a significant step away from the manual and varied practices currently in use, thereby aiding in monitoring response to therapies such as Targeted Radionuclide Therapy (TRT).

Implications and Future Work

One of the paper's main limitations is the presence of label noise in the public UniToChest dataset, which could affect the models’ performance accuracy. Addressing the dataset's label quality could further refine the results. Moreover, the lack of disease-specific information in the dataset restricts a more granular analysis of the models' clinical applicability across diverse lung cancer types or stages.

Looking to the future, implementing such AI-driven models into routine clinical practice could revolutionize how radiologists manage and monitor lung cancer. Furthermore, continuous advancements in AI for medical imaging, along with enhanced datasets, could see significant improvements in detection accuracy and reliability.

Conclusion

This paper presents a thorough investigation into the efficacy of deep learning models for the segmentation of lung lesions in CT scans. It underscores the potential for AI tools in improving lung cancer management but also highlights challenges in data quality and the integration of such technologies into clinical workflows. Continuous research and development in this domain could lead to impactful shifts in therapeutic monitoring, ultimately enhancing patient care outcomes.