HALLI: A new database of histolological Acute Lymphoblasitc Leukemia images

Acute Lymphoblastic or Lymphoid Leukemia, despite being rare, requires careful diagnosis to enhance outcome and decrease the incidence of remission or minimal residual disease. Microscopy, i.e., imaging, of marrow, biopsy, or peripheral blood, is still an essential diagnostic step, albeit now supported by genotypic, immunohistochemical and other investigations. In the context of using artificial intelligence, the analysis of available datasets demonstrated the use of solely images that were not associated with or linked to any other anamnestic or diagnostic data. This master thesis presents a dataset of images, which exhibit less interpretative variability than the histopathological report. Said images are associated with other data, such as age and sex, which form the basis for an evaluation by means of artificial intelligence. The dataset of images and metadata presented in this thesis was collected at the haematology clinic in Azienda Ospedaliero Universitaria delle Marche, located in Torrette di Ancona. The images were captured using the Nano Eye Device for Digital Hematology microscope and categorised into cases (patients) and controls (subjects in remission from other blood-related disorders). The total of 16 participants (11 cases and 5 controls) reflects the rarity of the medical condition and aligns with the hospital’s catchment area. It is important to note that this dataset is the first and only one presenting metadata and images from all three diagnostic investigations conducted to look for leukemia, as determined by medical literature. In detail, the dataset includes one peripheral blood smear, 13 bone marrow aspirates, and one bone marrow biopsy of leukemic and healthy lymphoblasts. This differs from the literature dataset, which only collects one type of diagnostic investigation. This dataset is unprecedented in that it includes a set of artifacts - specifically nuclei of cells that were crushed during slide preparation and are therefore lacking cytoplasm. As these nuclei can be perceived as both healthy and diseased, they pose a threat to the reliability of the results. Therefore, they must be analysed as distinct from the other cells and considered as artifacts. Moreover, researchers can utilize the metadata of each patient within the data set to supplement the input information provided for the artificial intelligence. This will enhance the cell classification feature, aligned with the techniques used by medical experts. In the future, it may offer support for the patient’s journey through the identification and correlation of related information.