Automated detection and classification of tumor histotypes on dynamic PET imaging data through machine-learning driven voxel classification.

2-deoxy-2-fluorine-(18F)fluoro-d-glucose Positron Emission Tomography/Computed Tomography (18F-FDG-PET/CT) is widely used in oncology mainly for diagnosis and staging of various cancer types, including lung cancer, which is the most common cancer worldwide. Since histopathologic subtypes of lung cancer show different degree of 18F-FDG uptake, to date there are some diagnostic limits and uncertainties, hindering an 18F-FDG-PET-driven classification of histologic subtypes of lung cancers. On the other hand, since activated macrophages, neutrophils, fibroblasts and granulation tissues also show an increased 18F-FDG activity, infectious and/or inflammatory processes and post-surgical and post-radiation changes may cause false-positive results, especially for lymph-nodes assessment. Here we propose a model-free, machine-learning based algorithm for the automated classification of adenocarcinoma, the most common type of lung cancer, and other types of tumors. Input for the algorithm are dynamic acquisitions of PET data (dPET), providing for a spatially and temporally resolved characterization of the uptake kinetic. The algorithm consists in a trained Random Forest classifier which, relying contextually on several spatial and temporal features of 18F-FDG uptake, generates as an outcome probability maps allowing to distinguish adenocarcinoma from other lung histotype and to identify metastatic lymph-nodes, ultimately increasing the specificity of the technique. Its performance, evaluated on a dPET dataset of 19 patients affected by primary lung cancer, provides a probability 0.943 ± 0.090 for the detection of adenocarcinoma. The use of this algorithm will guarantee an automatic and more accurate localization and discrimination of tumors, also providing a powerful tool for detecting at which extent tumor has spread beyond a primary tumor into lymphatic system.

Feasibility study on the use of CMOS sensors as detectors in radioguided surgery with ß-- emitters.

Radioguided surgery (RGS) is a medical practice which thanks to a radiopharmaceutical tracer and a probe allows the surgeon to identify tumor residuals up to a millimetric resolution in real-time. The employment of ß- emitters, instead of γ or ß+, reduces background from healthy tissues, administered activity to the patient, and medical exposure. In a previous work the possibility of using a CMOS Imager (Aptina MT9V011), initially designed for visible light imaging, to detect ß- from 90Y or 90Sr sources has been established. Because of its possible application as counting probe in RGS, the performances of MT9V011 in clinical-like conditions were studied.1 Through horizontal scans on a collimated 90Sr source of different sizes (1, 3, 5, 7 mm), we have determined relationships between scan fit parameters and the source dimension, namely A quadratic correlation and a linear dependency of, respectively, signal integrated over scan interval, and maximum signal against source diameter, are determined. Horizontal scan measurements on a source, interposing collimators of different size, aim to determine relationships or correlations between scan fit parameters and source dimension. A quadratic correlation and a linear dependency of, respectively, signal integrated over scan interval, and maximum signal against source diameter are determined. In order to get closer to clinical conditions, agar-agar phantoms containing 90Y with different dimensions and activities were prepared. A 90Y phantom is characterized by a central spot and a ring all around, for simulating both signal (tumor) and background (surrounding healthy tissue). The relationship found between scan maximum and 90Sr source diameter is then exploited to extract the concentration ratio between spot and external ring of the 90Y phantom. This observable, defined as the ratio between the tumor and the nearby healthy tissues uptake simulates the Tumor-to-Non-tumor Ratio (TNR). With the aim of evaluating the sensor's ability to discriminate signal from background relying on the significance parameter, a further 90Y phantom, featuring a well-known and clinical-like activity will mimic the signal only condition. This result is used to extrapolate to different source sizes, after having estimated the background for various TNR. The obtained significance values suggest that the MT9V011 sensor is capable of distinguishing a signal from an estimated background, depending on the interplay among TNR, acquisition time and tumor diameter.

Correction to: Comparison of three-parameter kinetic model analysis to standard Patlak's analysis in 18F-FDG PET imaging of lung cancer patients.

Following publication of the original article [1], the authors flagged that the author affiliations detailed in the article are incorrect for the authors M. L. Calcagni and A. Giordano.

Comparison of three-parameter kinetic model analysis to standard Patlak's analysis in 18F-FDG PET imaging of lung cancer patients.

BACKGROUND: Patlak's graphical analysis can provide tracer net influx constant (Ki) with limitation of assuming irreversible tracer trapping, that is, release rate constant (kb) set to zero. We compared linear Patlak's analysis to non-linear three-compartment three-parameter kinetic model analysis (3P-KMA) providing Ki, kb, and fraction of free 18F-FDG in blood and interstitial volume (Vb). METHODS: Dynamic PET data of 21 lung cancer patients were retrospectively analyzed, yielding for each patient an 18F-FDG input function (IF) and a tissue time-activity curve. The former was fitted with a three-exponentially decreasing function, and the latter was fitted with an analytical formula involving the fitted IF data (11 data points, ranging 7.5-57.5 min post-injection). Bland-Altman analysis was used for Ki comparison between Patlak's analysis and 3P-KMA. Additionally, a three-compartment five-parameter KMA (5P-KMA) was implemented for comparison with Patlak's analysis and 3P-KMA. RESULTS: We found that 3P-KMA Ki was significantly greater than Patlak's Ki over the whole patient series, + 6.0% on average, with limits of agreement of ± 17.1% (95% confidence). Excluding 8 out of 21 patients with kb > 0 deleted this difference. A strong correlation was found between Ki ratio (=3P-KMA/Patlak) and kb (R = 0.801; P < 0.001). No significant difference in Ki was found between 3P-KMA versus 5P-KMA, and between 5P-KMA versus Patlak's analysis, with limits of agreement of ± 23.0 and ± 31.7% (95% confidence), respectively. CONCLUSIONS: Comparison between 3P-KMA and Patlak's analysis significantly showed that the latter underestimates Ki because it arbitrarily set kb to zero: the greater the kb value, the greater the Ki underestimation. This underestimation was not revealed when comparing 5P-KMA and Patlak's analysis. We suggest that further studies are warranted to investigate the 3P-KMA efficiency in various tissues showing greater 18F-FDG trapping reversibility than lung cancer lesions.