Convolutional Neural Networks for Segmentation of Pleural Mesothelioma: Analysis of Probability Map Thresholds (CALGB 30901, Alliance).

The purpose of this study was to evaluate the impact of probability map threshold on pleural mesothelioma (PM) tumor delineations generated using a convolutional neural network (CNN). One hundred eighty-six CT scans from 48 PM patients were segmented by a VGG16/U-Net CNN. A radiologist modified the contours generated at a 0.5 probability threshold. Percent difference of tumor volume and overlap using the Dice Similarity Coefficient (DSC) were compared between the reference standard provided by the radiologist and CNN outputs for thresholds ranging from 0.001 to 0.9. CNN-derived contours consistently yielded smaller tumor volumes than radiologist contours. Reducing the probability threshold from 0.5 to 0.01 decreased the absolute percent volume difference, on average, from 42.93% to 26.60%. Median and mean DSC ranged from 0.57 to 0.59, with a peak at a threshold of 0.2; no distinct threshold was found for percent volume difference. The CNN exhibited deficiencies with specific disease presentations, such as severe pleural effusion or disease in the pleural fissure. No single output threshold in the CNN probability maps was optimal for both tumor volume and DSC. This study emphasized the importance of considering both figures of merit when evaluating deep learning-based tumor segmentations across probability thresholds. This work underscores the need to simultaneously assess tumor volume and spatial overlap when evaluating CNN performance. While automated segmentations may yield comparable tumor volumes to that of the reference standard, the spatial region delineated by the CNN at a specific threshold is equally important.

Convolutional Neural Networks for Segmentation of Malignant Pleural Mesothelioma: Analysis of Probability Map Thresholds (CALGB 30901, Alliance).

Malignant pleural mesothelioma (MPM) is the most common form of malignant mesothelioma, with exposure to asbestos being the primary cause of the disease. To assess response to treatment, tumor measurements are acquired and evaluated based on a patient's longitudinal computed tomography (CT) scans. Tumor volume, however, is the more accurate metric for assessing tumor burden and response. Automated segmentation methods using deep learning can be employed to acquire volume, which otherwise is a tedious task performed manually. The deep learning-based tumor volume and contours can then be compared with a standard reference to assess the robustness of the automated segmentations. The purpose of this study was to evaluate the impact of probability map threshold on MPM tumor delineations generated using a convolutional neural network (CNN). Eighty-eight CT scans from 21 MPM patients were segmented by a VGG16/U-Net CNN. A radiologist modified the contours generated at a 0.5 probability threshold. Percent difference of tumor volume and overlap using the Dice Similarity Coefficient (DSC) were compared between the standard reference provided by the radiologist and CNN outputs for thresholds ranging from 0.001 to 0.9. CNN annotations consistently yielded smaller tumor volumes than radiologist contours. Reducing the probability threshold from 0.5 to 0.1 decreased the absolute percent volume difference, on average, from 43.96% to 24.18%. Median and mean DSC ranged from 0.58 to 0.60, with a peak at a threshold of 0.5; no distinct threshold was found for percent volume difference. The CNN exhibited deficiencies with specific disease presentations, such as severe pleural effusion or disease in the pleural fissure. No single output threshold in the CNN probability maps was optimal for both tumor volume and DSC. This study emphasized the importance of considering both figures of merit when evaluating deep learning-based tumor segmentations across probability thresholds. This work underscores the need to simultaneously assess tumor volume and spatial overlap when evaluating CNN performance. While automated segmentations may yield comparable tumor volumes to that of the reference standard, the spatial region delineated by the CNN at a specific threshold is equally important.

Computer-assisted Curie scoring for metaiodobenzylguanidine (MIBG) scans in patients with neuroblastoma.

BACKGROUND: Radiolabeled metaiodobenzylguanidine (MIBG) is sensitive and specific for detecting neuroblastoma. The extent of MIBG-avid disease is assessed using Curie scores. Although Curie scoring is prognostic in patients with high-risk neuroblastoma, there is no standardized method to assess the response of specific sites of disease over time. The goal of this study was to develop approaches for Curie scoring to facilitate the calculation of scores and comparison of specific sites on serial scans. PROCEDURE: We designed three semiautomated methods for determining Curie scores, each with increasing degrees of computer assistance. Method A was based on visual assessment and tallying of MIBG-avid lesions. For method B, scores were tabulated from a schematic that associated anatomic regions to MIBG-positive lesions. For method C, an anatomic mesh was used to mark MIBG-positive lesions with automatic assignment and tallying of scores. Five imaging physicians experienced in MIBG interpretation scored 38 scans using each method, and the feasibility and utility of the methods were assessed using surveys. RESULTS: There was good reliability between methods and observers. The user-interface methods required 57 to 110 seconds longer than the visual method. Imaging physicians indicated that it was useful that methods B and C enabled tracking of lesions. Imaging physicians preferred method B to method C because of its efficiency. CONCLUSIONS: We demonstrate the feasibility of semiautomated approaches for Curie score calculation. Although more time was needed for strategies B and C, the ability to track and document individual MIBG-positive lesions over time is a strength of these methods.

Computer-assisted staging of chronic rhinosinusitis correlates with symptoms.

BACKGROUND: The Lund-Mackay (LM) staging system for chronic rhinosinusitis (CRS) does not correlate with clinical parameters, likely due to its coarse scale. We developed a "Modified Lund Mackay" (MLM) system, which uses a three-dimensional (3D), computerized method to quantify the volume of mucosal inflammation in the sinuses, and sought to determine whether the MLM would correlate with symptoms and disease-specific quality of life. METHODS: We obtained Total Nasal Symptom Score (TNSS) and 22-item Sino-Nasal Outcome Test (SNOT-22) data from 55 adult subjects immediately prior to sinus imaging. The volume of each sinus occupied by mucosal inflammation was measured using MATLAB algorithms created using customized, image analysis software after manual outlining of each sinus. Linear regression was used to model the relationship between the MLM and the SNOT-22 and TNSS. Correlation between the LM and MLM was tested using Spearman's rank correlation coefficient. RESULTS: Adjusting for age, gender, and smoking, a higher symptom burden was associated with increased sinonasal inflammation as captured by the MLM (ß = 0.453, p < 0.013). As expected due to the differences in scales, the LM and MLM scores were significantly different (p < 0.011). No association between MLM and SNOT-22 scores was found. CONCLUSION: The MLM is one of the first imaging-based scoring systems that correlates with sinonasal symptoms. Further development of this custom software, including full automation and validation in larger samples, may yield a biomarker with great utility for both treatment of patients and outcomes assessment in clinical trials.

Quantitative ultrasound image analysis of axillary lymph node status in breast cancer patients.

PURPOSE: Ultrasonography has the potential to accurately stage breast cancer with automated analysis to detect axillary lymph node metastasis. The aim of this study was to develop and test automated quantitative ultrasound image analysis of axillary lymph nodes for breast cancer staging. METHODS: Following an IRB-approved HIPAA compliant protocol, ultrasound images of 90 breast cancer patients presenting for lymph node assessment were retrospectively collected. There were 51 node-positive and 39 node-negative patients, yielding images of 223 lymph nodes (109 positive for metastasis and 114 negative for metastasis). The analysis was completely automated apart from the manual indication of the approximate center of each lymph node. Mathematical descriptors of the nodes, which served as image-based biomarkers, were computer-extracted and input to a classifier for the task of distinguishing between positive (i.e., metastatic) and negative lymph nodes. The performance of this task was assessed using receiver operating characteristic (ROC) analysis with evaluation by-node and by-patient using the area under the ROC curve (AUC) as the performance metric. RESULTS: The AUC was 0.85 (standard error 0.03) for by-node evaluation when distinguishing between positive and negative lymph nodes. The AUC was 0.87 (0.04) for patient-based prognosis, i.e., assessing whether patients were lymph node-positive or lymph node-negative. CONCLUSION: Based on these classification results, we conclude that mathematical descriptors of sonographically imaged lymph nodes may be useful as prognostic biomarkers in breast cancer staging and demonstrate potential for predicting patient lymph node status.

Utilisation of a thoracic oncology database to capture radiological and pathological images for evaluation of response to chemotherapy in patients with malignant pleural mesothelioma.

OBJECTIVE: An area of need in cancer informatics is the ability to store images in a comprehensive database as part of translational cancer research. To meet this need, we have implemented a novel tandem database infrastructure that facilitates image storage and utilisation. BACKGROUND: We had previously implemented the Thoracic Oncology Program Database Project (TOPDP) database for our translational cancer research needs. While useful for many research endeavours, it is unable to store images, hence our need to implement an imaging database which could communicate easily with the TOPDP database. METHODS: The Thoracic Oncology Research Program (TORP) imaging database was designed using the Research Electronic Data Capture (REDCap) platform, which was developed by Vanderbilt University. To demonstrate proof of principle and evaluate utility, we performed a retrospective investigation into tumour response for malignant pleural mesothelioma (MPM) patients treated at the University of Chicago Medical Center with either of two analogous chemotherapy regimens and consented to at least one of two UCMC IRB protocols, 9571 and 13473A. RESULTS: A cohort of 22 MPM patients was identified using clinical data in the TOPDP database. After measurements were acquired, two representative CT images and 0-35 histological images per patient were successfully stored in the TORP database, along with clinical and demographic data. DISCUSSION: We implemented the TORP imaging database to be used in conjunction with our comprehensive TOPDP database. While it requires an additional effort to use two databases, our database infrastructure facilitates more comprehensive translational research. CONCLUSIONS: The investigation described herein demonstrates the successful implementation of this novel tandem imaging database infrastructure, as well as the potential utility of investigations enabled by it. The data model presented here can be utilised as the basis for further development of other larger, more streamlined databases in the future.

Astrocyte and macrophage regulation of YKL-40 expression and cellular response in neuroinflammation.

Numerous inflammatory conditions are associated with elevated YKL-40 expression by infiltrating macrophages. Thus, we were surprised to observe minimal macrophage and abundant astrocyte expression of YKL-40 in neuroinflammatory conditions. The aims of the current study were to better delineate this discrepancy, characterize the factors that regulate YKL-40 expression in macrophages and astrocytes and study whether YKL-40 expression correlates with cell morphology and/or activation state. In vitro, macrophages expressed high levels of YKL-40 that was induced by classical activation and inhibited by alternative activation. Cytokines released from macrophages induced YKL-40 transcription in astrocytes that was accompanied by morphological changes and altered astrocytic motility. Because coculturing of astrocytes and macrophages did not reverse this in vitro expression pattern, additional components of the in vivo central nervous system (CNS) milieu must be required to suppress macrophage and induce astrocyte expression of YKL-40.

The Lung Image Database Consortium (LIDC) and Image Database Resource Initiative (IDRI): a completed reference database of lung nodules on CT scans.

PURPOSE: The development of computer-aided diagnostic (CAD) methods for lung nodule detection, classification, and quantitative assessment can be facilitated through a well-characterized repository of computed tomography (CT) scans. The Lung Image Database Consortium (LIDC) and Image Database Resource Initiative (IDRI) completed such a database, establishing a publicly available reference for the medical imaging research community. Initiated by the National Cancer Institute (NCI), further advanced by the Foundation for the National Institutes of Health (FNIH), and accompanied by the Food and Drug Administration (FDA) through active participation, this public-private partnership demonstrates the success of a consortium founded on a consensus-based process. METHODS: Seven academic centers and eight medical imaging companies collaborated to identify, address, and resolve challenging organizational, technical, and clinical issues to provide a solid foundation for a robust database. The LIDC/IDRI Database contains 1018 cases, each of which includes images from a clinical thoracic CT scan and an associated XML file that records the results of a two-phase image annotation process performed by four experienced thoracic radiologists. In the initial blinded-read phase, each radiologist independently reviewed each CT scan and marked lesions belonging to one of three categories ("nodule > or =3 mm," "nodule <3 mm," and "non-nodule > or =3 mm"). In the subsequent unblinded-read phase, each radiologist independently reviewed their own marks along with the anonymized marks of the three other radiologists to render a final opinion. The goal of this process was to identify as completely as possible all lung nodules in each CT scan without requiring forced consensus. RESULTS: The Database contains 7371 lesions marked "nodule" by at least one radiologist. 2669 of these lesions were marked "nodule > or =3 mm" by at least one radiologist, of which 928 (34.7%) received such marks from all four radiologists. These 2669 lesions include nodule outlines and subjective nodule characteristic ratings. CONCLUSIONS: The LIDC/IDRI Database is expected to provide an essential medical imaging research resource to spur CAD development, validation, and dissemination in clinical practice.

Characterization of mesothelioma and tissues present in contrast-enhanced thoracic CT scans.

PURPOSE: The purpose of this study was to characterize the Hounsfield unit (HU) distributions of mesothelioma and other tissues present in contrast-enhanced thoracic CT scans, to compare the HU distributions of mesothelioma, muscle, and liver by scanner and reconstruction filter/kernel combination, and to assess interpatient HU distribution variability. METHODS: The database consisted of 28 contrast-enhanced thoracic CT scans from different patients. For each scan, regions of interest were manually outlined within each of 13 tissues, including mesothelioma. For each tissue, the empirical percentiles in HU values were calculated along with the interpatient variability. The HU distributions of select tissues were compared across three different scanner and reconstruction filter/kernel combinations. RESULTS: The HU distributions of blood-containing tissues demonstrated substantial overlap, as did the HU distributions of pleural effusion, mesothelioma, muscle, and liver. The HU distribution of fat had the least overlap with the other tissues. Fat and muscle had the lowest interpatient HU variability and the narrowest HU distributions, while blood-containing tissues had the highest interpatient HU variability. A soft-tissue reconstruction filter/kernel yielded the narrowest HU distribution, and fat with artifact had the widest HU distribution. CONCLUSIONS: Characterization of tissues in CT scans enhances the understanding of those tissues' HU distributions. Due to their overlapping HU distributions and close spatial proximity to one another, separating pleural effusion, mesothelioma, muscle, and liver from one another is a difficult task based on HU value thresholding alone. The results illustrate the wide distributions and large variability that exist for tissues present in clinical thoracic CT scans.

Computerized segmentation and measurement of malignant pleural mesothelioma.

PURPOSE: The current linear method to track tumor progression and evaluate treatment efficacy is insufficient for malignant pleural mesothelioma (MPM). A volumetric method for tumor measurement could improve the evaluation of novel treatments, but a fully manual implementation of volume measurement is too tedious and time-consuming. This manuscript presents a computerized method for the three-dimensional segmentation and volumetric analysis of MPM. METHODS: The computerized MPM segmentation method segments the lung parenchyma and hemithoracic cavities to define the pleural space. Nonlinear diffusion and a k-means classifier are then implemented to identify MPM in the pleural space. A database of 31 computed tomography scans from 31 patients with pathologically confirmed MPM was retrospectively collected. Three observers independently outlined five randomly selected sections in each scan. The Jaccard similarity coefficient (J) between each of the observers and between the observer-defined and computer-defined segmentations was calculated. The computer-defined and the observer-defined segmentation areas (averaged over all observers) were both calculated for each axial section and compared using Bland-Altman plots. RESULTS: The median J value among observers averaged over all sections was 0.517. The median J between the computer-defined and manual segmentations was 0.484. The difference between these values was not statistically significant. The area delineated by the computerized method demonstrated variability and bias comparable to the tumor area calculated from manual delineations. CONCLUSIONS: A computerized method for segmentation and measurement of MPM was developed. This method requires minimal initialization by the user and demonstrated good agreement with manually drawn outlines and area measurements. This method will allow volumetric tracking of tumor progression and may improve the evaluation of novel MPM treatments.

Quantitative measurement of lung reexpansion in malignant pleural mesothelioma patients undergoing pleurectomy/decortication.

RATIONALE AND OBJECTIVES: Malignant pleural mesothelioma (MPM) is a neoplasm that grows circumferentially along the pleura. The tumor and concurrent pleural effusion may reduce lung function by restricting or preventing lung expansion. The purpose of this study was to provide objective evidence that pleurectomy/decortication (P/D) allows trapped lung to reexpand, quantify the reexpansion based on computed tomography (CT) scans, and investigate whether the expansion persists after surgery. MATERIALS AND METHODS: A database of 12 patients demonstrating unilateral MPM was collected. Each patient underwent a presurgical CT scan, surgical debulking by P/D, and two postsurgical CT scans (at 1 and 4 months). The lung volume was measured in each scan using an automated algorithm and compared for each patient across time. RESULTS: An increase in the ipsilateral postsurgical lung volume was observed for 10 of 12 patients (83%) 1 month after surgery. The median ipsilateral volume increase was 44% relative to the presurgical ipsilateral volume and 21% relative to the contralateral volume. A statistically significant change in ipsilateral lung volume was not observed between 1­month and 4­month postsurgical scans, implying that the volume improvement persisted months after surgery. CONCLUSIONS: Debulking of MPM with P/D substantially increased the ipsilateral lung volume relative to both the presurgical ipsilateral volume and the contralateral lung volume. This improvement persisted months after surgery.

In vivo CHI3L1 (YKL-40) expression in astrocytes in acute and chronic neurological diseases.

BACKGROUND: CHI3L1 (YKL-40) is up-regulated in a variety of inflammatory conditions and cancers. We have previously reported elevated CHI3L1 concentration in the cerebrospinal fluid (CSF) of human and non-human primates with lentiviral encephalitis and using immunohistochemistry showed that CHI3L1 was associated with astrocytes. METHODS: In the current study CHI3L1 transcription and expression were evaluated in a variety of acute and chronic human neurological diseases. RESULTS: ELISA revealed significant elevation of CHI3L1 in the CSF of multiple sclerosis (MS) patients as well as mild elevation with aging. In situ hybridization (ISH) showed CHI3L1 transcription mostly associated with reactive astrocytes, that was more pronounced in inflammatory conditions like lentiviral encephalitis and MS. Comparison of CHI3L1 expression in different stages of brain infarction showed that YKL40 was abundantly expressed in astrocytes during acute phases and diminished to low levels in chronic infarcts. CONCLUSIONS: Taken together, these findings demonstrate that CHI3L1 is induced in astrocytes in a variety of neurological diseases but that it is most abundantly associated with astrocytes in regions of inflammatory cells.

The influence of initial outlines on manual segmentation.

PURPOSE: Initial outlines are often presented as an aid to reduce the time-cost associated with manual segmentation and measurement of structures in medical images. This study evaluated the influence of initial outlines on manual segmentation intraobserver and interobserver precision. METHODS: Three observers independently outlined all pleural mesothelioma tumors present in five computed tomography (CT) sections in each of 30 patient scans. After a lapse of time, each observer was presented with the same series of CT sections with the outlines of each observer superimposed as initial outlines. Each observer created altered outlines by altering the initial outlines to reflect their perception of the tumor boundary. Altered outlines were compared to original outlines using the Jaccard similarity coefficient (J). Intraobserver and interobserver precision of observer outlines were calculated by applying linear mixed effects analysis of variance models to the J values. The percent of minor alterations (alterations that resulted in only slight changes in the initial outline) was also recorded. RESULTS: The average J value between pairs of observer original outlines was 0.371. The average J value between pairs of observer outlines when altered from an identical initial outline was 0.796, indicating increased interobserver precision. The average difference between J values of an observer's segmentation created by altering their own initial outline and when altering a different observer's initial outline was 0.476, indicating initial outlines strongly influence intraobserver precision. Observers made minor alterations on 74.5% of initial outlines with which they were presented. CONCLUSIONS: Intraobserver and interobserver precision were strongly dependent on the initial outline. These effects are likely due to the tendency of observers to make only minor corrections to initial outlines. This finding could impact observer study design, tumor growth assessment, computer-aided diagnosis system validation, and radiation therapy target volume definition when initial outlines are used as an observer aid.

YKL-40 expression in traumatic brain injury: an initial analysis.

YKL-40 (chitinase 3-like protein 1) is expressed in a broad spectrum of inflammatory conditions and cancers. We have previously reported that YKL-40 levels are elevated in the cerebrospinal fluid (CSF) of macaques and humans with lentiviral encephalitis, as well as multiple sclerosis (MS). The current study assessed temporal CSF YKL-40 levels in subjects with severe traumatic brain injury (TBI; Glasgow Coma Scale [GCS] score

Discrete-space versus continuous-space lesion boundary and area definitions.

Measurement of the size of anatomic regions of interest in medical images is used to diagnose disease, track growth, and evaluate response to therapy. The discrete nature of medical images allows for both continuous and discrete definitions of region boundary. These definitions may, in turn, support several methods of area calculation that give substantially different quantitative values. This study investigated several boundary definitions (e.g., continuous polygon, internal discrete, and external discrete) and area calculation methods (pixel counting and Green's theorem). These methods were applied to three separate databases: A synthetic image database, the Lung Image Database Consortium database of lung nodules and a database of adrenal gland outlines. Average percent differences in area on the order of 20% were found among the different methods applied to the clinical databases. These results support the idea that inconsistent application of region boundary definition and area calculation may substantially impact measurement accuracy.

Systemic and brain macrophage infections in relation to the development of simian immunodeficiency virus encephalitis.

The brains of individuals with lentiviral-associated encephalitis contain an abundance of infected and activated macrophages. It has been hypothesized that encephalitis develops when increased numbers of infected monocytes traffic into the central nervous system (CNS) during the end stages of immunosuppression. The relationships between the infection of brain and systemic macrophages and circulating monocytes and the development of lentiviral encephalitis are unknown. We longitudinally examined the extent of monocyte/macrophage infection in blood and lymph nodes of pigtailed macaques that did or did not develop simian immunodeficiency virus encephalitis (SIVE). Compared to levels in macaques that did not develop SIVE, more ex vivo virus production was detected from monocyte-derived macrophages and nonadherent peripheral blood mononuclear cells (PBMCs) from macaques that did develop SIVE. Prior to death, there was an increase in the number of circulating PBMCs following a rise in cerebrospinal fluid viral load in macaques that did develop SIVE but not in nonencephalitic macaques. At necropsy, macaques with SIVE had more infected macrophages in peripheral organs, with the exception of lymph nodes. T cells and NK cells with cytotoxic potential were more abundant in brains with encephalitis; however, T-cell and NK-cell infiltration in SIVE and human immunodeficiency virus encephalitis was more modest than that observed in classical acute herpes simplex virus encephalitis. These findings support the hypothesis that inherent differences in host systemic and CNS monocyte/macrophage viral production are associated with the development of encephalitis.

Evaluation of lung MDCT nodule annotation across radiologists and methods.

RATIONALE AND OBJECTIVES: Integral to the mission of the National Institutes of Health-sponsored Lung Imaging Database Consortium is the accurate definition of the spatial location of pulmonary nodules. Because the majority of small lung nodules are not resected, a reference standard from histopathology is generally unavailable. Thus assessing the source of variability in defining the spatial location of lung nodules by expert radiologists using different software tools as an alternative form of truth is necessary. MATERIALS AND METHODS: The relative differences in performance of six radiologists each applying three annotation methods to the task of defining the spatial extent of 23 different lung nodules were evaluated. The variability of radiologists' spatial definitions for a nodule was measured using both volumes and probability maps (p-map). Results were analyzed using a linear mixed-effects model that included nested random effects. RESULTS: Across the combination of all nodules, volume and p-map model parameters were found to be significant at P < .05 for all methods, all radiologists, and all second-order interactions except one. The radiologist and methods variables accounted for 15% and 3.5% of the total p-map variance, respectively, and 40.4% and 31.1% of the total volume variance, respectively. CONCLUSION: Radiologists represent the major source of variance as compared with drawing tools independent of drawing metric used. Although the random noise component is larger for the p-map analysis than for volume estimation, the p-map analysis appears to have more power to detect differences in radiologist-method combinations. The standard deviation of the volume measurement task appears to be proportional to nodule volume.

Variability in mesothelioma tumor response classification.

OBJECTIVE: The objective of our study was to evaluate observer variability in the measurement of temporal change in mesothelioma tumor thickness and in the resulting tumor response classification from CT scans. In addition, the performance of a semiautomated measurement method was evaluated. MATERIALS AND METHODS: Four observers individually used an interface that displayed two serial CT scans from the same patient to measure mesothelioma tumor thickness on the follow-up CT scans of 22 patients based on baseline scan measurements. During one session, observers acquired measurements on the follow-up scans based on written reports of baseline scan measurements; in another session, baseline scan measurements were superimposed on the baseline scan for direct visual comparison. Follow-up scan measurements also were obtained from a semiautomated method. Measurement variability and tumor response classification concordance were evaluated for manual measurements acquired in both modes and for semiautomated measurements. RESULTS: Although only a small increase in tumor response classification concordance rate was obtained with the visual approach (84.8%) relative to the more standard written-report approach (82.6%), the actual measurements acquired by observers were statistically significantly different between the two approaches (p = 0.03). Both the semiautomated measurements and the resulting tumor response classifications were consistent with manual measurements. CONCLUSION: The presentation of baseline scan tumor measurements affects measurements acquired on follow-up scans and could influence tumor response classification. The potential utility of semiautomated tumor thickness measurements was shown in the context of measuring tumor response.

Measurement of mesothelioma on thoracic CT scans: a comparison of manual and computer-assisted techniques.

Our purpose in this study was to evaluate the variability of manual mesothelioma tumor thickness measurements in computed tomography (CT) scans and to assess the relative performance of six computerized measurement algorithms. The CT scans of 22 patients with malignant pleural mesothelioma were collected. In each scan, an initial observer identified up to three sites in each of three CT sections at which tumor thickness measurements were to be made. At each site, five observers manually measured tumor thickness through a computer interface. Three observers repeated these measurements during three separate sessions. Inter- and intra-observer variability in the manual measurement of tumor thickness was assessed. Six automated measurement algorithms were developed based on the geometric relationship between a specified measurement site and the automatically extracted lung regions. Computer-generated measurements were compared with manual measurements. The tumor thickness measurements of different observers were highly correlated (r > or = 0.99); however, the 95% limits of agreement for relative inter-observer difference spanned a range of 30%. Tumor thickness measurements generated by the computer algorithms also correlated highly with the average of observer measurements (r > or = 0.93). We have developed computerized techniques for the measurement of mesothelioma tumor thickness in CT scans. These techniques achieved varying levels of agreement with measurements made by human observers.

Automated matching of temporally sequential CT sections.

In the evaluation of patient response to therapy through measurements on thoracic computed tomography (CT) scans, the selection of anatomically equivalent sections in temporally sequential scans is required. We developed an automated method based on normalized mutual information (NMI) to expedite the selection of anatomically equivalent sections. The method requires as input two temporally sequential CT scans from the same patient. A specified section from the baseline scan is then compared with the sections of a follow-up scan. Each section in the follow-up scan is successively translated and rotated relative to the baseline section, and NMI is calculated. The section in the follow-up scan that yields the highest NMI with respect to the baseline section is selected as the matching section. The method was applied to a database of 22 pairs of temporally sequential CT scans from mesothelioma patients. Five observers manually selected their choice of the best anatomically matched section for each of three predetermined sections in the 22 baseline scans, and the range of selected sections was recorded. The automated method was applied to the same baseline sections to determine the computer-based anatomically matched sections in the corresponding follow-up scan. The automated process was performed using both original CT sections and sections automatically segmented so that only intrathoracic pixels contributed to NMI calculations. The accuracy of the automated method was quantified on a section-by-section basis by comparison with the range of sections selected by the observers. The automated method without segmentation selected equivalent sections within the observers' range for 54 of the 66 matching tasks (81.8%). An 11% improvement was achieved when thoracic segmentation was performed as a pre-processing step.