Tensor radiomics: paradigm for systematic incorporation of multi-flavoured radiomics features

Background

Radiomics features hold significant value as quantitative imaging biomarkers for diagnosis, prognosis, and treatment response assessment. To generate radiomics features and ultimately develop signatures, various factors can be manipulated, including image discretization parameters (e.g., bin number or size), convolutional filters, segmentation perturbation, or multi-modality fusion levels. Typically, only one set of parameters is employed, resulting in a single value or "flavour" for each radiomics feature. In contrast, we propose "tensor radiomics" (TR) where tensors of features calculated using multiple parameter combinations (i.e., flavours) are utilized to optimize the creation of radiomics signatures.

Methods

We provide illustrative instances of TR implementation in positron emission tomography-computed tomography (PET-CT), magnetic resonance imaging (MRI), and CT by leveraging machine learning (ML) and deep learning (DL) methodologies, as well as reproducibility analyses: (I) to predict overall survival (OS) in lung cancer (CT) and head and neck cancer (PET-CT), TR was employed by varying bin sizes. This approach involved use of a hybrid deep neural network called 'TR-Net' and two ML-based techniques for combining different flavours. (II) TR was constructed by incorporating different segmentation perturbations and various bin sizes to classify the response of late-stage lung cancer to first-line immunotherapy using CT images. (III) In MRI of glioblastoma (GBM), TR was implemented to generate multi-flavour radiomics features, enabling enhanced analysis and interpretation. (IV) TR was employed via multiple PET-CT fusions in head and neck cancer. Flavours based on different fusions were created using Laplacian pyramids and wavelet transforms.

Results

Our findings demonstrated that TR outperformed conventional radiomics features in lung cancer CT and head and neck cancer PET-CT images, significantly enhancing OS prediction accuracy. TR also improved classification of lung cancer response to therapy and exhibited notable advantages in reproducibility compared to single-flavour features in MR imaging of GBM. Moreover, in head and neck cancer, TR through multiple PET-CT fusions exhibited improved performance in predicting OS.

Conclusions

We conclude that the proposed TR paradigm has significant potential to improve performance in different medical imaging tasks. By incorporating multiple flavours of radiomics features, TR overcomes limitations associated with individual features and shows promise in enhancing prognostic capabilities in clinical settings.

Prediction of Disease Progression to Upfront Pembrolizumab Monotherapy in Advanced Non-Small-Cell Lung Cancer with High PD-L1 Expression Using Baseline CT Disease Quantification and Smoking Pack Years

Health Canada approved pembrolizumab in the first-line setting for advanced non-small-cell lung cancer with PD-L1 ≥ 50% and no EGFR/ALK aberration. The keynote 024 trial showed 55% of such patients progress with pembrolizumab monotherapy. We propose that the combination of baseline CT and clinical factors can help identify those patients who may progress. In 138 eligible patients from our institution, we retrospectively collected their baseline variables, including baseline CT findings (primary lung tumor size and metastatic site), smoking pack years, performance status, tumor pathology, and demographics. The treatment response was assessed via RECIST 1.1 using the baseline and first follow-up CT. Associations between the baseline variables and progressive disease (PD) were tested by logistic regression analyses. The results showed 46/138 patients had PD. The baseline CT "number of involved organs" by metastasis and smoking pack years were independently associated with PD (p < 0.05), and the ROC analysis showed a good performance of the model that integrated these variables in predicting PD (AUC: 0.79). This pilot study suggests that the combination of baseline CT disease and smoking PY can identify who may progress on pembrolizumab monotherapy and can potentially facilitate decision-making for the optimal first-line treatment in the high PD-L1 cohort.

Self-reported vaccination against SARS-CoV-2 and adverse events in multiple cohorts

 

In two studies (“App-based infection assessment in RESIST (iAR)” and “Digital infection monitoring in persons living with immunodeficiency (DIMI)” ), we monitor health related items, as vaccination against SARS-CoV-2 and conduct syndromic surveillance of acute respiratory infections in high-risk populations, i.e. elderly persons and persons living with HIV, respectively. In a third very similar study (“Sensors for measuring aerosols and reactive gases to deduce health effects (SMARAGD)”) mainly healthy adults participate. To record incident or recurring transient health events, risk factors and further health data in real-time, we developed the eResearch system “PIA - Prospective Monitoring and Management App”. Recruitment for RESIST, SMARAGD and DIMI started in March 2021 and is ongoing. The questionnaire was presented in April 2022. Preliminary results include 86 participants from the three cohorts. In total, one indicated to be not vaccinated, none were vaccinated once, three (3.5%) twice, 63 (73.3%) three times and 19 (22.1%) four times. Participants reported the following adverse events after immunization (AEFI): after 40 applied doses with Vaxzevria® 24 AEFI (60%); after 158 doses of Comirnaty® 41 AEFI (26%); after 62 doses of Spikevax® 19 AEFI (30.7%); and after three doses of Janssen®, one AEFI (33.3%). In these cohorts, 20 (23.36%) participants stated having had a SARS-CoV-2 infection, of these 16 (80%) after the last vaccination dose, three (15%) before the first dose and one (5%) in between doses. Most participants were vaccinated three times, with Comirnaty being the most applied vaccine, as in officially reported numbers. AEFI varied according to vaccine and were higher than in the German surveillance system (1.64/1000 doses). Most infections were indicated to have been diagnosed after the booster vaccination. The results are limited by the small sample size and possible bias through self-reporting and social desirability regarding vaccination status.

Key messages

• Overall, most participants were vaccinated with Comirnaty and had three doses of vaccine. Of the participants with a diagnosed SARS-CoV-2-infection, most got infected after the booster vaccine.

• The number of reported AEFI was higher than in the official surveillance in Germany.

Abstract PO-053: Machine learning CADx process for classification of lung nodules below the Lung-RADS 4A threshold in LDCT scans

Lung Cancer screening trials have demonstrated significant mortality reduction. Low-Dose Computed Tomography (LDCT) screening can frequently discover many small nodules in at risk participants. However classification of these, sub-cm nodules as cancerous or benign is a challenging task even for expert clinicians. In this work we use machine learning (ML) techniques to differentiate, cancerous (clinically confirmed) and benign nodules (&gt;5 years of follow-up). Data for this study is drawn from a screening study (PanCan) from which we selected 613 distinct nodules (141 cancerous, and ~size matched 472 benign). We analyzed texture and shape features (~170) that are extracted from the nodule with and without perimeter transition pixels to control for perimeter effects. Features are also extracted from the ring of parenchyma surrounding the nodule to account for tumor effects on surrounding tissue. From the equivalent location in the opposite lung, parenchymal characteristics were extracted which we use to normalize the nodule texture features in an effort to reduce scanner bias. Our preliminary results for machine learning classification have shown model accuracies of up to ~80% (feature selection and classification algorithm dependent). Radiomic feature data can be combined with patient demographic variables such as age, sex, and smoking status to further improve our models, reaching ~84% classification accuracy. When normalized by the opposite lung, 58% of texture features showed improved classification ability. A single feature from the ring of parenchyma surrounding the nodule achieved 73.5% accuracy. Others have shown that nodule area/volume is a good classifier; for this data set it gave 68% accuracy. Data exploration to study classification accuracies among different subsets of patients reveals a number of interesting trends. We have noted for instance that in patients without emphysema, tumors can be classified with much higher accuracy (&gt;90%) than in those who suffer from it. This and other results detailing classification among specific patient groupings may prove relevant in any potential future clinical implementation of these results.

Citation Format: Rohan Abraham, Ian Janzen, Saeed Seyyedi, Sukhinder Khattra, John Mayo, Ren Yuan, Renelle Meyers, Stephen Lam, Calum MacAulay. Machine learning CADx process for classification of lung nodules below the Lung-RADS 4A threshold in LDCT scans [abstract]. In: Proceedings of the AACR Virtual Special Conference on Artificial Intelligence, Diagnosis, and Imaging; 2021 Jan 13-14. Philadelphia (PA): AACR; Clin Cancer Res 2021;27(5_Suppl):Abstract nr PO-053.

Submillimeter diameter rotary-pullback fiber-optic endoscope for narrowband red-green-blue reflectance, optical coherence tomography, and autofluorescence in vivo imaging

A fiber-based endoscopic imaging system combining narrowband red-green-blue (RGB) reflectance with optical coherence tomography (OCT) and autofluorescence imaging (AFI) has been developed. The system uses a submillimeter diameter rotary-pullback double-clad fiber imaging catheter for sample illumination and detection. The imaging capabilities of each modality are presented and demonstrated with images of a multicolored card, fingerprints, and tongue mucosa. Broadband imaging, which was done to compare with narrowband sources, revealed better contrast but worse color consistency compared with narrowband RGB reflectance. The measured resolution of the endoscopic system is 25  μm in both the rotary direction and the pullback direction. OCT can be performed simultaneously with either narrowband RGB reflectance imaging or AFI.

Advantages of short-lived positron-emitting radioisotopes for intracoronary radiation therapy with liquid-filled balloons to prevent restenosis

Balloon catheters filled with liquid radioisotopes provide excellent dose homogeneity for intracoronary radiation therapy but are associated with risk for rupture or leakage. We hypothesized that the safety of liquid-filled balloons may be improved once positron emitters with half-lives below 2 h are used instead of the high-energy beta-emitters 166Ho, 186Re, or 188Re, all of which have a longer half-life of at least 17 h.

METHODS

To support this concept, the suitability of 18F (half-life, 109.8 min), 68Ga (half-life, 67.6 min), 11C (half-life, 20.4 min), 13N (half-life, 9.97 min), and 15O (half-life, 2.04 min) for intracoronary radiation therapy was evaluated. Potential tissue penetration of positron radiation was assessed in a series of phantom experiments using Gafchromic film. Antiproliferative efficacy of positrons emitted by 68Ga was investigated in vitro using cultured bovine aortic smooth muscle cells (BASMCs), and was compared with gamma-radiation emitted by 137Cs. To characterize the remaining risk, we estimated radiotoxicity after accidental intravascular balloon rupture on the basis of tabulated isotope-specific doses (ICRP 53) and compared these values with 188Re.

RESULTS

Half-dose depth of tissue penetration measured in phantom experiments was 0.29 mm for 18F, 0.42 mm for 11C, 0.54 mm for 13N, 0.79 mm for 15O, and 0.9 mm for 68Ga. Irradiation of cultured BASMCs with positron radiation (68Ga) induced dose-dependent inhibition of proliferation with complete proliferative arrest at doses exceeding 6 Gy. ED(50) and ED(80) were 2.5 +/- 0.4 Gy (mean +/- SD) and 4.4 +/- 0.8 Gy, respectively. Antiproliferative efficacy was equal to that of the 662-keV gamma-radiation emitted by 137Cs (ED(50), 3.8 +/- 0.2 Gy; ED(80), 8.0 +/- 0.3 Gy). Estimates made for patient whole-body and organ doses were generally below 50 mSv/1.85 GBq for all investigated positron emitters. The same dose estimates for 188Re were 6-20 fold higher.

CONCLUSION

Among the studied radioisotopes, 68Ga is the most attractive source for liquid-filled balloons because of its convenient half-life, sufficient positron energy (2.92 MeV), documented antiproliferative efficacy, and uncomplicated availability from a radioisotope generator. The safety profile for 68Ga is significantly better than that of 188Re, which suggests this radioisotope should be evaluated further in preclinical studies.

[Heart enlargement in an athlete--a diagnostic challenge]

Endurance training can result in an enlargement of the heart. These athlete's hearts are rarer than generally assumed. Pathological causes, resulting in an eccentric hypertrophy, have to be considered. We report on a 32 year old athlete performing approximately 10 hours of endurance training weekly. He consulted a physician because of a drop in performance. The eccentric left ventricular hypertrophy, diagnosed by means of echocardiography, was not interpreted as a solely physiological cardiac adaptation because the ejection fraction did not increase during exercise (stress-echocardiography), the left ventricular diastolic function (maximum E/A-ratio) was impaired at rest, and the ergometric performance was reduced in comparison to the heart size. The invasive diagnostics including myocardial biopsy demonstrate histologically a focal fibrosis as the result of former myocarditis. The fibrosis was possibly involved in the genesis of the eccentric hypertrophy based on structural dilatation through a preferably mesenchymal lesion. It remains open whether the long-term endurance training had forced the dilatation. This case demonstrates that pathological causes must be excluded if in athletes an enlarged heart does not concur together with a clearly increased ergometric performance. Stress-echocardiography and endomyocardial biopsy can considerably contribute to the differential diagnosis between physiological and pathological cardiac hypertrophy.

Maternal concerns: what are they?

If one strives to provide care for the mother as a person adjusting to a major life change, then one needs to know more about the particular concerns of mothers of newborns. Open-ended interviews with mothers of newborns provided a list of basic concerns, which was then used to create a self-administered check-list. Eighty-four mothers of newborns revealed that their top three concerns were "feeling exhausted," "amount of time as a couple," and "body weight." These concerns were similar for many subgroups of mothers except when the babies were older than 10 weeks of age. Paradoxically, physicians and nurses providing post-partum care are not seen as sources of help for any of the 10 most frequent concerns reported by the mothers.

[Ulceration of the skin in necrobiosis lipoidica (author's transl)]

As a supplement to earlier investigations, a case of a 46 year old female patient is reported who has suffered from necrobiosis lipoidica granulomatosa with ulceration for 17 years. A marked remission was obtained by treatment with anti-inflammatory and immunosuppressive therapy. It is pointed out that under the usual treatment when ulcers of the lower leg and other areas of the skin do not heal, after exclusion of malignancies, necrobiosis lipoidica diabeticorum or necrobiosis lipoidica granulomatosa must be considered, especially as prophylactic examinations may arise.