Development of machine learning methods for multi-modal biomarkers detection and integration

In medicine, innovation depends on a better knowledge of the human body mechanism, which represents a complex system of multi-scale constituents. Unraveling the complexity underneath diseases proves to be challenging. A deep understanding of the inner workings comes with dealing with many heterogeneous information. Exploring the molecular status and the organization of genes, proteins, metabolites provides insights on what is driving a disease, from aggressiveness to curability. Molecular constituents, however, are only the building blocks of the human body and cannot currently tell the whole story of diseases. This is why nowadays attention is growing towards the contemporary exploitation of multi-scale information. Holistic methods are then drawing interest to address the problem of integrating heterogeneous data. The heterogeneity may derive from the diversity across data types and from the diversity within diseases. Here, four studies conducted data integration using customly designed workflows that implement novel methods and views to tackle the heterogeneous characterization of diseases. The first study devoted to determine shared gene regulatory signatures for onco-hematology and it showed partial co-regulation across blood-related diseases. The second study focused on Acute Myeloid Leukemia and refined the unsupervised integration of genomic alterations, which turned out to better resemble clinical practice. In the third study, network integration for artherosclerosis demonstrated, as a proof of concept, the impact of network intelligibility when it comes to model heterogeneous data, which showed to accelerate the identification of new potential pharmaceutical targets. Lastly, the fourth study introduced a new method to integrate multiple data types in a unique latent heterogeneous-representation that facilitated the selection of important data types to predict the tumour stage of invasive ductal carcinoma. The results of these four studies laid the groundwork to ease the detection of new biomarkers ultimately beneficial to medical practice and to the ever-growing field of Personalized Medicine.

Anatomical biomarkers of body composition in oncological population: application and development of artificial intelligences

In the Era of precision medicine and big medical data sharing, it is necessary to solve the work-flow of digital radiological big data in a productive and effective way. In particular, nowadays, it is possible to extract information “hidden” in digital images, in order to create diagnostic algorithms helping clinicians to set up more personalized therapies, which are in particular targets of modern oncological medicine. Digital images generated by the patient have a “texture” structure that is not visible but encrypted; it is “hidden” because it cannot be recognized by sight alone. Thanks to artificial intelligence, pre- and post-processing software and generation of mathematical calculation algorithms, we could perform a classification based on non-visible data contained in radiological images. Being able to calculate the volume of tissue body composition could lead to creating clasterized classes of patients inserted in standard morphological reference tables, based on human anatomy distinguished by gender and age, and maybe in future also by race. Furthermore, the branch of “morpho-radiology" is a useful modality to solve problems regarding personalized therapies, which is particularly needed in the oncological field. Actually oncological therapies are no longer based on generic drugs but on target personalized therapy. The lack of gender and age therapies table could be filled thanks to morpho-radiology data analysis application.

Combining molecular alterations and functional imaging in metastatic castration resistant prostate cancer treated with taxanes

The treatment of metastatic castration-resistant prostate cancer (mCRPC) is currently characterized by several drugs with different mechanisms of action, such as new generation hormonal agents (abiraterone, enzalutamide), chemotherapy (docetaxel, cabazitaxel), PARP inhibitors (olaparib) and radiometabolic therapies (radium-223, LuPSMA). There is an urgent need to identify biomarkers to guide personalized therapy in mCRPC. In recent years, the status of androgen receptor (AR) gene detected in liquid biopsy has been associated with outcomes in patients treated with abiraterone or enzalutamide. More recently, plasma tumor DNA (ptDNA) and its changes during treatment have been identified as early indicators of response to anticancer treatments. Recent works also suggested a potential role of tumor-related metabolic parameters of 18Fluoro-Choline Positron Emission Tomography (F18CH-PET)-computed tomography (CT) as a prognostic tool in mCRCP. Other clinical features, such as the presence of visceral metastases, have been correlated with outcome in mCRPC patients. Recent studies conducted by our research group have designed and validated a prognostic model based on the combination of molecular characteristics (ptDNA levels), metabolic features found in basal FCH PET scans (metabolic tumor volume values, MTV), clinical parameters (absence or presence of visceral metastases), and laboratory tests (serum lactate dehydrogenase levels, LDH). Within this PhD project, 30 patients affected by mCRPC, pre-treated with abiraterone or enzalutamide, candidate for taxane-based treatments (docetaxel or cabazitaxel), have been prospectively evaluated. The prognostic model previously described was applied to this population, to interrogate its prognostic power in a more advanced cohort of patients, resulting in a further external validation of the tool.

Response times in computerized adaptive testing: a method for cheating detection

In the field of educational and psychological measurement, the shift from paper-based to computerized tests has become a prominent trend in recent years. Computerized tests allow for more complex and personalized test administration procedures, like Computerized Adaptive Testing (CAT). CAT, following the Item Response Theory (IRT) models, dynamically generates tests based on test-taker responses, driven by complex statistical algorithms. Even if CAT structures are complex, they are flexible and convenient, but concerns about test security should be addressed. Frequent item administration can lead to item exposure and cheating, necessitating preventive and diagnostic measures. In this thesis a method called "CHeater identification using Interim Person fit Statistic" (CHIPS) is developed, designed to identify and limit cheaters in real-time during test administration. CHIPS utilizes response times (RTs) to calculate an Interim Person fit Statistic (IPS), allowing for on-the-fly intervention using a more secret item bank. Also, a slight modification is proposed to overcome situations with constant speed, called Modified-CHIPS (M-CHIPS). A simulation study assesses CHIPS, highlighting its effectiveness in identifying and controlling cheaters. However, it reveals limitations when cheaters possess all correct answers. The M-CHIPS overcame this limitation. Furthermore, the method has shown not to be influenced by the cheaters’ ability distribution or the level of correlation between ability and speed of test-takers. Finally, the method has demonstrated flexibility for the choice of significance level and the transition from fixed-length tests to variable-length ones. The thesis discusses potential applications, including the suitability of the method for multiple-choice tests, assumptions about RT distribution and level of item pre-knowledge. Also limitations are discussed to explore future developments such as different RT distributions, unusual honest respondent behaviors, and field testing in real-world scenarios. In summary, CHIPS and M-CHIPS offer real-time cheating detection in CAT, enhancing test security and ability estimation while not penalizing test respondents.

Quantitative susceptibility mapping as biomarker of neurodegeneration: methodological models and clinical applications

Quantitative Susceptibility Mapping (QSM) is an advanced magnetic resonance technique that can quantify in vivo biomarkers of pathology, such as alteration in iron and myelin concentration. It allows for the comparison of magnetic susceptibility properties within and between different subject groups. In this thesis, QSM acquisition and processing pipeline are discussed, together with clinical and methodological applications of QSM to neurodegeneration. In designing the studies, significant emphasis was placed on results reproducibility and interpretability. The first project focuses on the investigation of cortical regions in amyotrophic lateral sclerosis. By examining various histogram susceptibility properties, a pattern of increased iron content was revealed in patients with amyotrophic lateral sclerosis compared to controls and other neurodegenerative disorders. Moreover, there was a correlation between susceptibility and upper motor neuron impairment, particularly in patients experiencing rapid disease progression. Similarly, in the second application, QSM was used to examine cortical and sub-cortical areas in individuals with myotonic dystrophy type 1. The thalamus and brainstem were identified as structures of interest, with relevant correlations with clinical and laboratory data such as neurological evaluation and sleep records. In the third project, a robust pipeline for assessing radiomic susceptibility-based features reliability was implemented within a cohort of patients with multiple sclerosis and healthy controls. Lastly, a deep learning super-resolution model was applied to QSM images of healthy controls. The employed model demonstrated excellent generalization abilities and outperformed traditional up-sampling methods, without requiring a customized re-training. Across the three disorders investigated, it was evident that QSM is capable of distinguishing between patient groups and healthy controls while establishing correlations between imaging measurements and clinical data. These studies lay the foundation for future research, with the ultimate goal of achieving earlier and less invasive diagnoses of neurodegenerative disorders within the context of personalized medicine.

Cancers of unknown primary site (CUPs) as a model of metastatic process.

Cancers of unknown primary site (CUPs) are a rare group of metastatic tumours, with a frequency of 3-5%, with an overall survival of 6-10 month. The identification of tumour primary site is usually reached by a combination of diagnostic investigations and immunohistochemical testing of the tumour tissue. In CUP patients, these investigations are inconclusive. Since international guidelines for treatment are based on primary site indication, CUP treatment requires a blind approach. As a consequence, CUPs are usually empiric treated with poorly effective. In this study, we applied a set of microRNAs using EvaGreen-based Droplet Digital PCR in a retrospective and prospective collection of formalin-fixed paraffin-embedded tissue samples. We assessed miRNA expression of 155 samples including primary tumours (N=94), metastases of known origin (N=10) and metastases of unknown origin (N=50). Then, we applied the shrunken centroids predictive algorithm to obtain the CUP’s site(s)-of-origin. The molecular test was successfully applied to all CUP samples and provided a site-of-origin identification for all samples, potentially within a one-week time frame from sample inclusion. In the second part of the study we derived two CUP cell lines, and corresponding patient-derived xenografts (PDXs). CUP cell lines and PDXs underwent histological, molecular, and genomic characterization confirming the features of the original tumour. Tissues-of-origin prediction was obtained from the tumour microRNA expression profile and confirmed by single cell RNA sequencing. Genomic testing analysis identified FGFR2 amplification in both models. Drug-screening assays were performed to test the activity of FGFR2-targeting drug and the combination treatment with the MEK inhibitor trametinib, which proved to be synergic and exceptionally active, both in vitro and in vivo. In conclusion, our study demonstrated that miRNA expression profiling could be employed as diagnostic test. Then we successfully derived two CUP models from patients, used for therapy tests, bringing personalized therapy closer to CUP patients.

Development of a health parameters tracking wearable device for detection of stress episodes and principles of panic attacks

A proof of concept for a wearable device is presented to help patients who suffer from panic attacks due to panic disorder. The aim of this device is to enable such patients manage these stressful episodes by guiding them to regulate their breathing and by informing the care taker. Panic attack prediction is deployed that can enable the healthcare providers to not only monitor and manage the panic attacks of a patient but also carry out an early intervention to reduce the symptom severity of the approaching panic attack. The patient can acquire the help they need, ultimately regaining control. The concept of panic attack prediction can lead to a personalized treatment of the patient. The study is conducted using a small real-world dataset, and only two primary symptoms of panic attack are used. These symptoms include pacing heart rate and hyperventilation or abnormal breathing rate. This thesis project is developed in collaboration with ALTEN italia and all the required hardware is provided by them.