Machine learning: applications of artificial intelligence to imaging and diagnosis.

Machine learning (ML) is a form of artificial intelligence which is placed to transform the twenty-first century. Rapid, recent progress in its underlying architecture and algorithms and growth in the size of datasets have led to increasing computer competence across a range of fields. These include driving a vehicle, language translation, chatbots and beyond human performance at complex board games such as Go. Here, we review the fundamentals and algorithms behind machine learning and highlight specific approaches to learning and optimisation. We then summarise the applications of ML to medicine. In particular, we showcase recent diagnostic performances, and caveats, in the fields of dermatology, radiology, pathology and general microscopy.

Identification of a regulatory loop for the synthesis of neurosteroids: a steroidogenic acute regulatory protein-dependent mechanism involving hypothalamic-pituitary-gonadal axis receptors.

Brain sex steroids are derived from both peripheral (primarily gonadal) and local (neurosteroids) sources and are crucial for neurogenesis, neural differentiation and neural function. The mechanism(s) regulating the production of neurosteroids is not understood. To determine whether hypothalamic-pituitary-gonadal axis components previously detected in the extra-hypothalamic brain comprise a feedback loop to regulate neuro-sex steroid (NSS) production, we assessed dynamic changes in expression patterns of steroidogenic acute regulatory (StAR) protein, a key regulator of steroidogenesis, and key hypothalamic-pituitary-gonadal endocrine receptors, by modulating peripheral sex hormone levels in female mice. Ovariectomy (OVX; high serum gonadotropins, low serum sex steroids) had a differential effect on StAR protein levels in the extrahypothalamic brain; increasing the 30- and 32-kDa variants but decreasing the 37-kDa variant and is indicative of cholesterol transport into mitochondria for steroidogenesis. Treatment of OVX animals with E(2), P(4), or E(2) + P(4) for 3 days, which decreases OVX-induced increases in GnRH/gonadotropin production, reversed this pattern. Suppression of gonadotropin levels in OVX mice using the GnRH agonist leuprolide acetate inhibited the processing of the 37-kDa StAR protein into the 30-kDa StAR protein, confirming that the differential processing of brain StAR protein is regulated by gonadotropins. OVX dramatically suppressed extra-hypothalamic brain gonadotropin-releasing hormone 1 receptor expression, and was further suppressed in E(2)- or P(4)-treated OVX mice. Together, these data indicate the existence of endocrine and autocrine/paracrine feedback loops that regulate NSS synthesis. Further delineation of these feedback loops that regulate NSS production will aid in developing therapies to maintain brain sex steroid levels and cognition.

Copper Induces Apoptosis of Neuroblastoma Cells Via Post-translational Regulation of the Expression of Bcl-2-family Proteins and the tx Mouse is a Better Model of Hepatic than Brain Cu Toxicity.

The basic mechanism(s) by which altered Cu homeostasis is toxic to hepatocytes and neurons, the two major cell types affected in copper storage diseases such as Wilson's disease (WD), remain unclear. Using human M17 neuroblastoma cells as a model to examine Cu toxicity, we found that there was a time- and concentration-dependent induction of neuronal death, such that at 24 h there was a approximately 50 % reduction in viability with 25 muM Cu-glycine(2). Cu-glycine(2) (25:50 muM) treatment for 24 h significantly altered the expression of 296 genes, including 8 genes involved with apoptosis (BCL2-associated athanogene 3, BCL2/adenovirus E1B 19kDa interacting protein caspase 5, regulator of Fas-induced apoptosis, V-jun sarcoma virus 17 oncogene homolog, claudin 5, prostaglandin E receptor 3 and protein tyrosine phosphatase, non-receptor type 6). Surprisingly, changes in the expression of more 'traditional' apoptotic genes (Bcl-2, Bax, Bak and Bad) did not vary more than 20 %. To test whether the induction of apoptosis in neuroblastoma cells was via post-translational mechanisms, we measured the protein expression of these apoptotic markers in M17 neuroblastoma cells treated with Cu-glycine(2) (0-100 muM) for 24-48 h. Compared with glycine treated cells, Cu-glycine(2) reduced Bcl-2 expression by 50 %, but increased Bax and Bak expression by 130% and 400 %, respectively. To assess whether Cu also induced apoptotic cell death in a mouse model of WD, we measured the expression of these apoptotic markers in the liver and brain of mice expressing an ATP7b gene mutation (tx(J) mice) at 10 months of age (near the end of their lives when overt liver pathology is displayed). Changes in the liver expression of these apoptotic markers in tx(J) mice compared to background mice mirrored those of Cu treated neuroblastoma cells. In contrast, few changes in apoptotic protein expression were detected in the brain between tx(J) and background mice, indicating the tx(J) mouse is a good model of hepatic, but not brain, Cu toxicity. Our results indicate that Cu-induction of neuronal apoptosis does not require de novo synthesis or degradation of apoptotic genes, and that Cu accumulation in the aged tx(J) mouse brain is insufficient to induce apoptosis.

Amyloid-beta-induced toxicity of primary neurons is dependent upon differentiation-associated increases in tau and cyclin-dependent kinase 5 expression.

It has previously been reported that amyloid-beta (Abeta) peptide is neurotrophic to undifferentiated but neurotoxic to differentiated primary neurons. The underlying reasons for this differential effect is not understood. Recently, the toxicity of Abeta to neurons was shown to be dependent upon the activation of cyclin-dependent kinase 5 (Cdk5), thought to promote tau phosphorylation that leads to cytoskeletal disruption, morphological degeneration and apoptosis. Here we report that Cdk5, tau, and phosphorylated-tau (P-tau) are expressed at very low levels in undifferentiated primary neurons, but that the expression of Cdk5 and tau and the phosphorylation of tau increase markedly between 4 and 8 days of differentiation in vitro. Tau expression decreased after this time, as did the level of P-tau, to low levels by 17 days. Abeta induced tau phosphorylation of neurons only after >or= 4 days of differentiation, a time that coincides with the onset of Abeta toxicity. Blocking tau expression (and therefore tau phosphorylation) with an antisense oligonucleotide completely blocked Abeta toxicity of differentiated primary neurons, thereby confirming that tau was essential for mediating Abeta toxicity. Our results demonstrate that differentiation-associated changes in tau and Cdk-5 modulate the toxicity of Abeta and explain the opposite responses of differentiated and undifferentiated neurons to Abeta. Our results predict that only cells containing appreciable levels of tau are susceptible to Abeta-induced toxicity and may explain why Abeta is more toxic to neurons compared with other cell types.